Wayne State UniversityDigitalCommons@WayneState

English Faculty Research Publications English

8-1-2013

Accent in UspantekoRyan BennettYale University

Robert HendersonWayne State University, rhenderson@wayne.edu

This Article is brought to you for free and open access by the English at DigitalCommons@WayneState. It has been accepted for inclusion in EnglishFaculty Research Publications by an authorized administrator of DigitalCommons@WayneState.

Recommended CitationBennett, Ryan & Henderson, Robert. "Accent in Uspanteko." Natural Language & Linguistic Theory 31.3 (2013): 589-645. doi:10.1007/s11049-013-9196-6Available at: http://digitalcommons.wayne.edu/englishfrp/22

Accent in Uspanteko

Ryan Bennett∗ Robert Henderson†

Abstract

Uspanteko (Guatemala; ∼2000 speakers) is an endangered K’ichean-branch Mayan language. It isunique among the K’ichean languages in having innovated a system of contrastive pitch accent, whichoperates alongside a separate system of non-contrastive stress. The prosody of Uspanteko is of generaltypological interest, given the relative scarcity of ‘mixed’ languages employing both stress and lexicalpitch. Drawing from a descriptive grammar and from our own fieldwork, we also document some intricateinteractions between pitch accent and other aspects of the phonology (stress placement, vowel length,vowel quality, and two deletion processes). While pitch accent is closely tied to morphology, the locationof lexical tone is entirely a matter of surface phonology. We propose that the position of pitch accent andstress is determined by three factors: (i) feet are always right-aligned, and preferably iambic; (ii) pitchaccent must fall on a stressed syllable; and (iii) pitch accent cannot fall on a final mora. These assumptionsderive default final stress, as well as a regular pattern of tone-triggered stress shift. Interactions betweenprosody and segmental phonotactics are attributed to further constraints on footing. Surprisingly, wefind robust evidence for foot structure in Uspanteko, even though these patterns could easily be describedin non-metrical terms. Interactions between tone and vowel length also provide evidence for lexical stratawithin the Uspanteko vocabulary.

1 Introduction

Uspanteko is a Mayan language spoken in and around the municipality of Uspantan (Tz’unun Kaab’) in thedepartment of Quiche in the western highlands of Guatemala. While the K’ichean-branch as a whole is largeand robust, Uspanteko is severely threatened.

Uspanteko

K’icheanproper

KaqchikelTz’utujil

K’ichee’K’ichean Achi

SakapultekoSipakapense

Poqom PoqomchiPoqomam

Q’eqchi’

Figure 1: K’ichean-branch Mayan languages (after Kaufman 1974; Richards 2003)

∗Yale University†Wayne State University

1

Though there have probably never been many Uspanteko speakers, there are now only about 1500–2000(Richards 2003), and they are for the most part trilingual, speaking both K’ichee’ and Spanish as well(Can Pixabaj 2006). With these considerations in mind, Uspanteko can be classified as endangered.

Figure 2: Map of Guatemala showing Uspanteko-speaking region (adapted from Icke 2007)

Uspanteko is also underdocumented. Currently available materials include a descriptive grammar anddictionary, both published recently by Oxlajuuj Keej Maya’ Ajtz’iib’ (OKMA) (Can Pixabaj 2006; Mendez2007). Campbell (1977) also discusses the language briefly in his reconstruction of Proto-K’ichean. Despitethese excellent resources, the phonology is still largely underdescribed (at least relative to the morphosyntaxof the language, which is better understood). The paucity of phonological description for Uspanteko is acritical gap because it is the only language in the K’ichean-branch that makes use of lexical tone.

Building on these extant resources with our own original fieldwork, this paper pushes forward our un-derstanding of Uspanteko stress and tone. Contrary to earlier work that ascribes a binary tonal inventoryto Uspanteko (Grimes 1971; Campbell 1977; Can Pixabaj 2006), we show that Uspanteko has a single Htone that is restricted to the penultimate mora of the word. Since default stress in Uspanteko is word-final,and tone and stress must co-occur on the same syllable, tone realization sometimes requires stress shift. Weargue that this tone-driven stress shift should be treated as an iambic-trochaic foot-form reversal. We thenanalyze a series of complex, and otherwise puzzling tone-segment and tone-morphology interactions throughthe simple combination of footing constraints and constraints barring final H tone. One interesting outcomeis that different stems behave differently with respect to the interaction of tone and vowel length, which wetake to motivate distinct cophonologies in the nominal domain.

The overarching goal of this paper is to provide a fairly comprehensive description and analysis of word-level prosody in Uspanteko. While pursuing that core aim, we will also argue that the accentual systemof Uspanteko has a number of theoretically and typologically interesting properties, despite the surfacesimplicity of the system itself. Since this is the first in-depth treatment of Uspanteko prosody, it necessarilytouches on a range of diverse phenomena. A secondary goal of this paper, then, is to show that Uspantekoprosody is also less complex than it might at first seem, in that a number of apparently disparate empiricalfacts can be explained with only a small set of assumptions about prosodic structure.

We begin by introducing the basic phonology of Uspanteko in Section 1.1. Section 2 extends this de-scription by detailing the properties of tone in Uspanteko. In Section 3 we provide an analysis of these basicfacts. Section 4.1 looks at a non-local tone blocking effect manifested by a particular type of coda cluster.Section 4.2 zooms in and considers cophonologies that require slight permutations of the analysis proposedin Section 3. Section 5 concludes.

1.1 Basic phonology

As is typical in Mayan languages, the morphophonology of Uspanteko is built around a set of CV(V)Croots. Complex words are mostly formed through suffixing derivational morphology and prefixing inflectional

2

morphology. There is also a large number of clitics in both the verbal and nominal domains. The word-levelprosodic integration of some of these clitics is discussed in the appendix. Barring the consideration of clitics,main word stress is final and there is no evidence of secondary stress.

All examples in this paper are presented in standard Mayan orthography.1 The Mayan orthographicsystem is largely phonemic, with the following departures from the IPA consonant system:

IPA Standard MayanVelar fricative /x/ jPalato-alveolar fricative /S/ xPalatal approximant /j/ y

Palato-alveolar affricate />tS/ ch

Alveolar affricate />ts/ tz

Glottal stop /P/ ’Glottalized consonants /á/, /kP/, etc. b’, k’, etc.

Figure 3: Divergences between Mayan orthography and IPA transcription

The phonemic consonants of Uspanteko are given in Figure 4 (in standard Mayan orthography). Thereare no examples of a glottal stop directly following another consonant in this paper, so [C’] unambiguouslyindicates a glottalized consonant rather than a consonant-glottal stop sequence.

Bilabial Alveolar Palato-alveolar Palatal Velar Uvular Glottal

Stop p b’ t t’ k k’ q q’ ’Affricate tz tz’ ch ch’Fricative s x jSonorant m w n l r y

Figure 4: Uspanteko consonant phonemes

The vowel inventory of Uspanteko consists of short [i e a o u] and their long counterparts. The orthographicrepresentation of vowels also corresponds fairly closely to their actual IPA values, with one important ex-ception: in unstressed syllables, orthographic short 〈a〉 is phonetically [@] (see Section 3.3).

2 Tone

This section considers the core properties of Uspanteko tone that we examine: namely, the basic tonalinventory, and the interaction of tone with segmental, suprasegmental, and morphological phenomena.

As far as we are aware, all previous work dealing with Uspanteko agrees that the language makes useof contrastive lexical tone (e.g. Grimes 1971; Campbell 1977; Can Pixabaj 2006). There is also a generalconsensus that word-level stress is obligatory, but lexical tone is not: all words in Uspanteko bear stress, butonly some words carry contrastive tone as well.

1Since we provide orthographic forms, a number of regular allophonic processes (e.g. nasal place assimilation, initial glottalstop insertion, etc.) are not represented in the data. Such processes are discussed where relevant.

3

(1) Stress is obligatory in Uspanteko; tone is not

a. [. . . "σ ]b. [. . . "σT ]

c. *[. . . σ ]d. *[. . . σT ]

To restate the generalization, there are no words in the language that bear lexical tone without also bearingstress (1-d), though there are many words with regular final stress and no independent specification ofword-level tone (1-a). On these broad points, we are in agreement with the existing literature on Uspanteko.

However, there are important differences between our understanding of Uspanteko tone and the viewsespoused in earlier research. Beginning with the tonal inventory, we depart from previous treatments ofUspanteko in assuming that all surface tones are the expression of a single, privative, high tone pitch accent.

(2) Uspanteko tone is a privative high tone

a. [. . . CVH .CV]b. [. . . CVHV]

This contrasts with Grimes (1971), Campbell (1977), and Can Pixabaj (2006), who each propose that thereare two distinct tones in Uspanteko. For example, Can Pixabaj (2006) claims that there is an H tone thatonly occurs on penultimate short vowels, as well as an L or falling tone that only occurs on final long vowels.

(3) Analysis of Uspanteko tone in Can Pixabaj (2006) (to be rejected)

a. Tone 1: [. . . CVH .CV]b. Tone 2: [. . . CVVL]

Note that the assumption of this two tone inventory for Uspanteko is not motivated by the actual tonalcontrasts found in the language. For one, the proposed H and L tones never contrast with each otherbecause they never appear in the same positions. They also do not co-occur: words of the shape [CV.CVV]can only bear tone on the final long vowel, and not on both syllables simultaneously. Before getting off theground, then, such an analysis misses the generalization that the two purported tones do not behave likeseparate tonemes: they are in complementary distribution, and they never co-occur.

Grimes (1971) and Campbell (1977) sketch a slightly different picture of tone in Uspanteko (thoughCampbell is quite clear that his analysis is a working hypothesis, and not a definitive claim). In their view,contrastive tone is limited to stressed, word-final long vowels. Stressed long vowels may bear either hightone or low tone; short vowels may not carry tone of any sort, even when stressed.

(4) Analysis of Uspanteko tone in Grimes (1971) and Campbell (1977) (to be rejected)

a. Tone 1: [. . . CVVH ]b. Tone 2: [. . . CVVL]

As we show below, this view is incorrect. Short vowels have higher pitch in stressed penults than in stressedword-final syllables. This position-dependent increase in pitch can be straightforwardly understood as thethe phonetic realization of a phonological H tone on stressed CV(C) penults, which is wholly unexpectedunder the assumption that only long vowels can bear lexical tone.

Given these difficulties, we prefer to posit just one underlying tone for Uspanteko, deriving any surfaceallophonic differences in how that tone is realized (i.e. as high vs. falling) from the properties of theenvironment it appears in. In line with this goal, we simplify the tonal inventory by assuming a single,privative H tone, as in (2) above. With this simplification, a better distributional generalization emerges:H tone appears as close to the end of the word as possible, while satisfying additional constraints on wheretone may be realized.

We take the tone-bearing unit (TBU) for Uspanteko to be the vocalic mora (µV ) (see Section 4.1 for whywe limit TBUs to vocalic moras alone). We further assume that the H tone of Uspanteko can only appear

4

on a non-word-final TBU (more on this in Section 3).2,3 Under these assumptions, H tone will surface onthe penult in words ending in a light CV(C) syllable: when the final vowel is short, the nucleus of the penultwill be the rightmost non-final TBU, and will thus host tone (if tone is present). In tonal words endingwith a heavy CVV(C) syllable, H will appear on the final syllable itself because tone can associate with thefirst mora of the long vowel, again the rightmost non-final TBU. Long vowels only appear word-finally inUspanteko, so these two structures exhaust the space of possible tonal configurations. (From here on [V]indicates a vocalic mora bearing high tone.)

(5) Uspanteko tone: privative H tone on rightmost non-final TBU

a. [. . . CVµ.CVµ]

b. [. . . CVµVµ]c. *[. . . CVµVµ. . . ]

We thus eliminate the L tone proposed by previous authors—which under the analysis of Can Pixabaj (2006)would be in complementary distribution with H at any rate—in favor of a single H tone along with conditionson where that tone can be realized. The appearance of tone on a final heavy syllable or a light penult isentirely determined by well-formedness conditions on where tone can appear, so the position of H tone doesnot need to be underlyingly specified.

If the sole tone in Uspanteko is indeed a high tone H, as we propose, the question arises as to whyprevious researchers posited a low or falling tone on final long vowels instead. We think that tone on longvowels may have been perceived as low or falling because the H target is on the first mora, and after that Htarget is reached pitch may fall through the rest of the vowel (see e.g. Myers 1998). In our own fieldworkwe’ve found no phonetic indication of distinct word-level L tones; and phonologically, the evidence clearlyfavors a single tone analysis.4

2.1 Tone-stress interactions

Now that we have a picture of the tonal inventory, we can begin to consider the interaction of tone and otherphonological processes, beginning with stress. Like other K’ichean-branch Mayan languages, default stressin Uspanteko is on the final syllable (from here onward, stress is indicated by underlining). This is truewhether the final vowel is long or short, and whether the final syllable is open CV(V) or closed CV(V)C.(Examples come from our original fieldwork, unless otherwise noted. Data cited from Can Pixabaj (2006)were verified by our consultants whenever possible.)

(6) a. [chikach] ‘basket’b. [xib’alb’al] ‘half-brother’c. [tiqab’ana’] ‘we’re doing it’d. [lajori] ‘today’e. [xinlowisaaj] ‘I sheparded it.’ (Can Pixabaj 2006:22,33,70–71,280)

Final stress is thus ‘default’ in two senses: syllable quantity does not affect where stress falls; and non-tonalwords have final stress without exception (we return to stress shift in tonal words shortly).

We use the term ‘stress’ primarily in its phonological sense: stressed syllables are syllables that, in virtueof their position within some word, bear abstract structural prominence. This abstract prominence may (butneed not) be phonetically expressed by increasing the relative acoustic salience of stressed syllables along

2Carlos Gussenhoven (p.c.) has suggested to us that the functional grounding of this constraint might be linked to thefact that Uspanteko has large rising pitch excursions at the ends of phrase-level intonational domains. The language-internalmotivation for barring lexical H from final moras, then, is the avoidance of tonal crowding at the right edge of phrases (e.g.Gordon 2000). See also ? for related discussion of phrase-final intonational targets in K’ichee’.

3The avoidance of domain-final high tones is well-attested crosslinguistically: for example, see Cassimjee and Kisseberth(2007) and Hyman (2007) on Bantu languages; Pulleyblank (1986) on Margi (Chadic); Kawahara and Shinya (2008) on Japanese;Demers et al. (1999:43) on Yaqui; Silverman and Pierrehumbert (1990) on English (at the phonetic level); various examples inYip (2002:29,66,90–1); and the general discussion in Hyman (1977).

4As for the high [. . . CVVH ] tone in Grimes (1971) and Campbell (1977), we suspect that those authors may have misin-terpreted phrase-level H tones—which also dock to final stressed syllables—as being instances of lexical tone. Alternatively,their high tone may be our tonal [. . . CVV], and their low tone our toneless [. . . CVV]. Given the very small number of examplewords provided by those authors, it’s difficult to determine which of these alternatives is more plausible.

5

dimensions like intensity, duration, etc. (see Fry 1955, Bollinger 1958, Liberman 1975, Liberman and Prince1977, Hayes 1995, and Cutler 2005 for extended discussion on these points). Two distributional facts pointtoward default final stress in Uspanteko. First, long vowels are permitted only in word-final syllables, asevidenced by length alternations like (7).

(7) No non-final long vowels in Uspanteko

a. [chuun] ‘lime (mineral)’b. [x-chun-aj] ‘he covered it with lime’ (Can Pixabaj 2006:53)

Since stressed syllables commonly host more phonological contrasts than unstressed syllables (e.g. Trubet-zkoy 1939; Beckman 1998), the fact that vowel length contrasts are limited to final position is a credibleindication that stress is final as well (especially given the cross-linguistic tendency to neutralize lengthcontrasts in final syllables, e.g. Buckley 1998; Barnes 2006; Myers and Hansen 2007).

Second, syllable codas consisting of a glottal stop and a following consonant are also restricted to finalposition.

(8) No non-final [CV’C] syllables in Uspanteko

a. [ka’n] ‘animal’b. [kuwa’y] ‘horse’c. *[wa’yku]

As we argue in Section 3 and 4.1, this is no coincidence: only [VV(C)] and [V’C] rimes count as heavyin Uspanteko, and heavy syllables are only allowed word-finally. This distributional restriction on syllabletypes follows naturally if default stress is word-final: like many languages, Uspanteko enforces an outrightban on unstressed heavy syllables; and given final stress, this has the effect of limiting heavy syllables tofinal position as well.

Finally, the phonologically exceptional status of final syllables in Uspanteko would be deeply puzzlingunder the assumption that such syllables are unstressed. Initial syllables are often positions of phonologicalprivilege, probably for psycholinguistic reasons concerning the role of initial syllables in lexical access (seee.g. Beckman 1998; Nelson 2003; Smith 2005). Final syllables are rarely (perhaps never) privileged in thesame way, unless some other phonological factor (like stress) renders final position prominent on independentgrounds (though cf. Barnes 2006). There are thus sound language-internal and typological motivations forassuming final stress for Uspanteko. Since Campbell (1977) and Can Pixabaj (2006) are in accord with thisview, we believe that the existence of default final stress in Uspanteko is beyond plausible doubt.5

Exceptions to default final stress are entirely systematic: non-final stress is found only when word-leveltone falls on the penultimate syllable. If a word bears pitch accent (i.e. H tone), tone and stress mustcoincide; and in words with a tonal penult (i.e. tonal words ending in a short vowel), the non-finalitycondition on tone placement takes precedence over default final stress, leading to stress shift.

(9) a. [walib’] ‘my daughter-in-law’b. [ab’aj] ‘stone’c. [mu’xix] ‘bellybutton’d. [ıwir] ‘yesterday’ (Can Pixabaj 2006:56)

In the remainder of the paper, the term ‘accented’ refers to the syllable (or vowel) bearing stress, or bearingboth stress and tone. When referring to tone in isolation, apart from stress, we will sometimes use the morespecific term ‘pitch accent’.

Before we derive these tone-stress interactions, we want to forestall an alternative analysis that takespenultimate stress placement to be a case of stress shift alone, lacking an independent tonal element. Weknow that penultimate H is not merely stress shift because stress and tone have separable phonetic correlates:penultimate H involves a pitch excursion above and beyond any pitch perturbations associated with finalstress. Compare the minimally different forms [intz’i’] ‘I am a dog’ (Figure 5) and [ıntz’i’] ‘my dog’ (Figure

5Note also that the closely-related Mayan languages K’ekchi, Kaqchikel (Berinstein 1979), K’ichee’ (Pye 1983), and Tz’utujil(Dayley 1985) uncontroversially have final stress as well.

6

6).6

30

55

80In

ten

sity

(d

B)

80

190

300

Pit

ch (

Hz)

38ms 53ms 101ms 105ms 188ms

i n tz’ i ’

Time (s)0 0.5721

Figure 5: Intensity, pitch, and duration for [intz’i’] ‘I am a dog’ (final stress, no tone)

Tonal [ıntz’i’] has a clear pitch peak on the accented syllable (the penult), whereas the stressed final syllableof non-tonal [intz’i’] lacks any such pitch excursion. For non-tonal [intz’i’], the ratio of mean pitch duringthe penultimate vowel versus the final vowel is 1.26 (difference: 38 Hz); the ratio of the pitch peaks duringthose vowels is 1.14 (difference: 23 Hz). For tonal [ıntz’i’], the pitch differences between the penult andultima are quite a bit larger: the ratio of means is 1.63 (difference: 100 Hz), and the ratio of peaks is 1.53(difference: 91 Hz). If penultimate accent were a simple case of stress shift, we might expect final stress toalso be correlated with raised pitch on the stressed vowel; instead, in Figure 5 we find uninterrupted pitchdeclination from the penult to the ultima. Penultimate accented vowels thus have an independent tonalcomponent not found on stressed final short vowels.

A reviewer expresses concern that the final glottal stop in non-tonal [intz’i’] ‘I am a dog’ might bedepressing pitch in the last syllable, thereby obscuring any tonal correlates of default final stress. But asFigures 7 and 8 show, stressed final syllables in non-tonal words lack appreciable pitch movement even whenending in a sonorant or non-glottalized obstruent. Compare these examples to tonal [lekej] ‘up’ (Figure 9),which has a strongly uneven pitch profile marked by a sharp rise during the penult. Pitch measurements forFigures 5–9 are given in Table 1.7

6The utterances shown in Figures 5–9 were produced by a female speaker of Uspanteko in her mid-thirties. This speakeris originally from Uspantan, though for the past few years she has been living in Duenas, a small town outside of Antiguain Guatemala. For ease of comparison, all phonetic diagrams presented in this paper correspond to productions by this samespeaker, recorded over a single three-day period in March 2011.

All tokens were elicited in a phrasal context. For example, we elicited [ıntz’i’] ‘my dog’ by asking the question (in Spanish)“Who saw your dog on Monday?”. This prompted a response like [Tek xril ıntz’i’ lunes] “Tek saw my dog on Monday”. Thismethod allowed us to elicit target words in phrase-medial position (thereby avoiding interference from phrase-final intonationalcontours) and in a discourse-given context (thereby avoiding potential interference from focus prosody on the word of interest).

7The same reviewer argues that the gently rising pitch contour on the final syllable of [tulul] (Figure 7) indicates thathigh pitch is in fact a correlate of stress in Uspanteko. While this may ultimately prove correct, it remains true that pitchperturbations on stressed penults are larger and steeper than corresponding pitch changes on stressed ultimas. It is thisasymmetry that motivates our claim that penultimate accent involves an additional phonological H tone.

7

30

55

80

Inte

nsi

ty (

dB

)

80

190

300

Pit

ch (

Hz)

127ms 87ms 96ms 105ms 133ms

í n tz’ i ’

Time (s)0 0.6243

Figure 6: Intensity, pitch, and duration for [ıntz’i’] ‘my dog’ (penultimate stress with tone)

WordMean vowel pitch (Hz) Peak vowel pitch (Hz)

Penult Final Diff. Ratio Penult Final Diff. Ratio

Non-tonal

[intz’i’] 187 149 38 1.26 192 169 23 1.14

[tulul] 209 212 3 0.99 221 224 3 0.99

[in-chiij] 220 211 9 1.04 223 214 9 1.04

Tonal[ıntz’i’] 259 159 100 1.63 262 171 91 1.53

[lekej] 266 189 77 1.41 280 193 87 1.45

Table 1: Relative pitch measurements for penultimate and final vowels in Figures 5–9

Having demonstrated that lexical pitch can be phonetically distinguished from stress, it remains to beshown that penultimate pitch accent also involves the retraction of stress from the final syllable to the penult.In non-tonal words, default final stress is often realized with some degree of vowel lengthening, relative tounstressed vowels of the same quality and phonemic length. For example, the stressed final [i] in non-tonal[intz’i’] ‘I am a dog’ (Figure 5) is about 2.75 times as long as the unstressed [i] in the penult (difference:67ms). Similarly, the stressed final [u] in non-tonal [tulul] ‘zapote (sp. of tree)’ (Figure 7) is about 1.7 timesas long as the unstressed [u] in the penult (difference: 38ms). We thus take it as a working hypothesis thatvowel lengthening is a correlate of stress in Uspanteko.

Now note that the nucleus of the accented, penultimate syllable in [ıntz’i’] (Figure 6) has over threetimes the duration of its unaccented counterpart in the penult of non-tonal [intz’i’] (Figure 5; difference:89ms). The penultimate, tone-bearing [ı] in [ıntz’i’] is also 22ms longer than the final [i] in the same word.Finally, in [lekej] (Figure 9) the penultimate, tone-bearing [e] is 2.4 times as long as the [e] in the last syllable(difference: 68ms). We conclude from all of this that the vowel lengthening found on tone-bearing penultsis in fact indicative of stress shift, as claimed by Can Pixabaj (2006:71).8

8If duration is indeed a correlate of stress in Uspanteko, we might ask why the final [i] in Figures 5 and 6 is 105ms long whetheror not it is stressed. While we have not yet conducted a full-scale study of the phonetics of stress in Uspanteko, we suspect that

8

30

55

80

Inte

nsi

ty (

dB

)

80

190

300

Pit

ch (

Hz)

118ms 55ms 106ms 93ms 125ms

t u l u l

Time (s)0 0.496

Figure 7: Intensity, pitch, and duration for [tulul] ‘zapote (sp. of tree)’ (final stress, no tone)

While we are still investigating the phonetic correlates of stress in Uspanteko, it is nevertheless clear thatstress and tone are dissociated, independent aspects of word-level accent in the language. We conclude, then,that the realization of tone on the penult does indeed draw stress away from its default final position, underpressure for tone and stress to coincide. A satisfactory analysis of Uspanteko accent must capture this fact.

2.2 Tone-segment interactions

Tone regularly interacts with segmental structure in Uspanteko. We find segments that block tone realization,as well as segments that are correlated with its appearance. This work focuses primarily on the first class,though we present an analysis of the latter class in Section 3.3.1.

The first generalization is that while long vowels can bear tone lexically, it is more generally the casethat long vowels block the realization of tone when introduced by other morphemes. For instance, somepossessive prefixes introduce an H tone that must be realized on the penult (10), but it is not normallyrealized on nominals with long vowels (11) (see also Figure 8).

(10) a. [tz’i’] ‘dog’b. [ın-tz’i’] ‘my dog’

c. [laq] ‘plate’d. [ın-laq] ‘my plate’

e. [teleb’] ‘shoulder’

duration is sometimes suppressed as a correlate of stress in word-final syllables, since vowel length is only contrastive in finalposition. See Berinstein (1979) for discussion of similar facts in the closely-related Mayan languages Kaqchikel and K’eqchi;and see Campos-Astorkiza (2007) for general discussion. Alternatively, it may be that duration is a general correlate of stressin Uspanteko, but the phonetic cues of stress are only weakly and irregularly realized, as in many languages with fixed stress(e.g. French and Czech, Cutler 2005; Bengali, Hayes and Lahiri 1991; Bininj Gun-Wok, Bishop 2002).

9

30

55

80

Inte

nsi

ty (

dB

)

80

190

300

Pit

ch (

Hz)

53ms 58ms 118ms 149ms 44ms

i n ch ii j

Time (s)0 0.4996

Figure 8: Intensity, pitch, and duration for [in-chiij] ‘my hair ribbon’ (final stress, no tone)

f. [inteleb’] ‘my shoulder’

(11) a. [in-chaa] ‘my obsidian’b. [in-b’aaq] ‘my bone’c. [in-b’iis] ‘my sadness’

The effect is not limited to long vowels. We see the same behavior with CV’C syllables ([CVPC] in IPAnotation).

(12) a. [in-ch’o’j] ‘my fight’b. [in-kuwa’y] ‘my horse’c. [in-ka’n] ‘my animal’

Especially interesting is the fact that the conditioning CV’C syllables are word-final, so they themselves arenot even potential hosts for the tone. We thus need an analysis that can explain why these syllables triggertone deletion at a distance.

While CV’C syllables block tone realization, there are other syllables that are correlated with tone.Can Pixabaj (2006) notes that a large number of bisyllabic words that have final [a]/[i] nuclei bear tone.It is important to recognize that /a/ reduces to [@] in unstressed syllables, which we transcribe here forexplicitness.

(13) a. [ıchij] ‘greens’b. [ıxim] ‘corn’c. [ajij] ‘cane’d. [xılik] ‘inside’e. [ıj@j] ‘seed’f. [ab’@j] ‘stone’g. [tun@q’] ‘Adam’s apple’

10

30

55

80

Inte

nsi

ty (

dB

)

80

190

300

Pit

ch (

Hz)

118ms 116ms 100ms 48ms 76ms

l é k e j

Time (s)0 0.5087

Figure 9: Intensity, pitch, and duration for [lekej] ‘up’ (penultimate stress with tone)

h. [pat@n] ‘burden’etc.

The basic pattern is that tone appears in bisyllabic forms when the final, and ultimately post-tonic syllablewould have relatively low sonority. Further examples provide evidence that the sonority profile of thesebisyllabic words is really what’s at issue. Tone is also very common for bisyllables with [a] in the penult—that is, tone appears in the vast majority of bisyllabic words where it would yield a tonic syllable withrelatively high sonority.

(14) a. [ak’el] ‘child’b. [chaqej] ‘dry’c. [ajche’] ‘matazano (species of tree)’

Finally, a large number of disyllabic words with vowels of equal sonority have tone.

(15) a. [etzel] ‘evil’b. [ojor] ‘a long time ago’c. [tukun] ‘blackberry’d. [lekej] ‘up’e. [tz’unun] ‘hummingbird’

Taking into consideration these extensions to Can Pixabaj’s (2006) generalization about final [a]/[i] in bi-syllabic words, the basic pattern is that tone is inserted (with stress shift) when the relative sonority of thetonic vowel with respect to the post-tonic vowel is no worse than it would be if the word did not bear tone.Of course, we have to make precise what it means for a word’s sonority profile to be ‘no worse’ under toneinsertion, as well as account for why this calculation is restricted to bisyllabic root words. While we wait tillSection 3.3.1 to build our account of these facts, what’s important to draw from these data is that, just aswith syllables that block tone, the properties of non-local segments matter for conditioning tone.

11

2.3 Tone-morphology interactions

Finally, in addition to tones specified on lexical words, there are certain functional morphemes that introducean H tone. Certain possessive prefixes, for instance, are associated with the appearance of tone (10).However, since H has a restricted distribution, morphemes introducing tone often do not bear that tonethemselves.

(16) -ib’ plural

a. [ajk’ay] ‘seller’b. [ajk’ay-ib’] ‘sellers’

c. [ajchaak] ‘worker’d. [ajchak-ib]’ ‘workers’

(17) -wu VP focus clitic

a. [kla’ xintıj-wu] ‘It was there where I ate.’b. [lamaas wı-wu] ‘Where did you eat?’

(18) Local person possessive prefixes9

a. [ixk’eq] ‘fingernail’b. [w-ıxk’eq] ‘my fingernail’

c. [teleb’] ‘shoulder’d. [in-teleb’] ‘my shoulder’

(19) Phrase final status suffix -ik10

a. [xinchakun . . . ] ‘I worked . . . ’b. [xinchakun-ik] ‘I worked.’

c. [xinel . . . ] ‘I left . . . ’d. [xinel-ik] ‘I left.’

The possessive prefixes are particularly interesting because they also serve as ergative agreement markers,but do not trigger tone when crossreferencing the subject of a transitive verb.

(20) a. [t-in-loq’]inc-E1s-buy‘I bought it.’

b. *[t-ın-loq’]

The conclusion is that the H tone is associated with an abstract morpheme implicated in agreement withgenitive nominals, and not simply part of the phonological spellout of the ergative prefixes in general. A nicemorphological consequence is that we can use H tone as a probe for true nominal possession. For instance,Mayan languages are known for marking oblique relations using what are called relational nouns. These arebound morphemes that crossreference their arguments with a single ergative agreement morpheme, exactlylike possessed nouns crossreference their possessors. While relational nouns look like possessed nominals, wecan show through the distribution of tone that they are not in fact possessed.

(21) [-ik’il] ‘with’

a. [w-ik’il] ‘with me’b. *[wık’il]

9Third person possessive prefixes do not introduce tone, a fact that we will return to.10Status suffixes indicate verb class membership, where classes are defined in terms of TAM, transitivity, and whether the

verb is root or derived (see Kaufman 1990; ? for a discussion of status suffixes in the closely related language K’ichee’).

12

(22) [-ib’] reflexive

a. [aw-ib’] ‘yourself’b. *[awib’]

3 Analysis

3.1 Sources of tone

With the basic facts laid out in the previous section, we now build an analysis of Uspanteko tone. Sincethere is only one toneme, namely H, and its distribution is predictable, we assume that words bearing tonein the output contain a free-floating H tone associated with some morpheme in the input.11

(23) a. /anim, H/ ‘woman’ → [anim]b. /ajchaak + -ib’, H/ ‘workers’ → [ajchak-ib’]

(Can Pixabaj 2006:58,62)

The eventual placement of H is determined entirely by surface phonological constraints, so there is noempirical reason to assume that tone is ever linked to a particular position in the underlying representation.

While some instances of H tone need to be specified in the lexicon, there are certain cases where tonepredictably appears across a morpho-syntactic paradigm. For example, possession is realized through acombination of prefixal ergative marking and tone realization.

(24) Tone and possessive marking

a. [aqan] ‘leg’b. [w-aqan] ‘my leg’c. [aw-aqan] ‘your leg’ (Can Pixabaj 2006:27,54,546)

11See Yip (2002:Section 9.3), Gussenhoven (2004:Section 3.5), and Kubozono (2008:171) on related ‘diacritic’ approaches totone and pitch accent.

13

d. [pix] ‘tomato’e. [ın-pix] ‘my tomato’f. [qa-pix] ‘our tomato’g. [a-pix=aq] ‘your (pl.) tomato’12

In the absence of tone examples like (24) are either ungrammatical, or can only be interpreted as a segmentallyhomophonous non-verbal predication structure (with an absolutive rather than an ergative prefix).13

(25) Non-verbal predication lacks tone

a. [ın-kar]erg.1s-fish‘my fish’

b. [in-kar]abs.1s-fish‘I am a fish.’ (Can Pixabaj 2006:64)

c. [ın-jal]erg.1s-ear.of.corn‘my ear of corn’

d. [in-jal]abs.1s-ear.of.corn‘I am an ear of corn.’

One analytical option is to assume that the ergative prefixes are themselves specified with tone in the input.The problem with this approach is that we miss the generalization that all and only those ergative prefixesappearing in possessive constructions bear tone—verbal ergative markers are always non-tonal.

(26) Verbal ergative marking lacks tone

a. [x-in-qej]asp.comp-erg.1s-lend‘I lent it.’ (Can Pixabaj 2006:42)

b. *[x-ın-qej]

c. [k-oj-a-chap-e’]imp-abs.1p-erg.2s-grab-imp‘Grab us!’ (Can Pixabaj 2006:226)

d. *[k-oj-a-chap-e’]

Assuming an underlying tonal specification for the ergative prefixes also leads to non-trivial redundancy. Eachergative prefix has multiple phonologically-conditioned allomorphs: for example, the first-person singularergative possessive prefix is [in-] before consonants (e.g. [ın-chi’] ‘my mouth’), but normally [w-] beforevowels (e.g. [w-ıxk’eq] ‘my fingernails’) (Can Pixabaj 2006:57,92). Any account of Uspanteko possessivesthat assumes tone is underlyingly associated to the ergative prefixes must then posit tone on each individualergative allomorph—thereby reducing a systematic fact about tone distribution to the level of a lexicalaccident.

Instead, we propose that tone is the spell-out of the syntactic head responsible for assigning genitive Caseto possessors. We call this head F, and remain non-committal with respect to its actual syntactic category.We further assume that possessive constructions like (24) have the following syntactic structure:

(27) Basic syntax of possessive constructions in Uspanteko

12On the plural clitic [=aq], see the appendix and example (97).13There are cases of possession in which tone fails to surface for phonological reasons. These cases are discussed and analyzed

in later sections.

14

FP

(DPposs)

F

H

NP

Tone, then, is simply the spell-out of the F head that assigns genitive Case in possessive constructions.14

(Note that both full DP and null possessors are allowed in Uspanteko; see e.g. Aissen 1999 for relateddiscussion.)

A further complication arises when we consider that morphological tone only appears on possessed nom-inals if their possessor is first- or second-person.15

(28) No tone triggered by third person possessors16

a. [kaa’] ‘grinding stone’

b. [ın-ki’] ‘my grinding stone’c. [qa-ki’] ‘our grinding stone’

d. [a-ki’] ‘your grinding stone’e. [a-ki’=aq] ‘your (pl.) grinding stone’

f. [j-kaa’] ‘his/her grinding stone’g. *[j-kaa’]

h. [j-kaa’=aq] ‘their grinding stone’i. *[j-kaa’=aq]

We assume that first- and second-person possessors—so-called local person possessors—are syntacticallydistinguished from third-person possessors by virtue of bearing the feature specification [+participant] (seee.g. Nevins 2007 and references therein). We make the additional assumption that when F assigns genitiveCase to a possessor, F takes on the φ-features of that possessor, including its person features. A feature-copying mechanism of this sort is fundamental to much recent work in Minimalist syntax (e.g. Chomsky2001), though our account departs from standard forms of Minimalism in assuming that feature-copyinghappens in a spec-head configuration in possessives. In the presence of a local person possessor, then, Fwill come to bear the feature [+participant] as well. The distribution of tone in possessive configurationscan then be formalized as in (29), using notation familiar from work in Distributed Morphology (DM; Halleand Marantz 1993, 1994; Harley and Noyer 1999; Embick and Noyer 2007).

(29) Insertion rules for head F

a. F, [+participant] ↔ Hb. F ↔ Ø

These rules of vocabulary insertion (along with an elsewhere condition like the subset principle; e.g. Halleand Marantz 1993, 1994; Embick and Halle 2005, etc.) guarantee that possession-triggered tone will onlysurface in the presence of a local person possessor.

What of the ergative possessive prefixes? While providing a full account of the syntax of possession inUspanteko is well beyond the scope of this paper, we would nonetheless like to speculate on the syntacticrealization of the ergative prefixes. Following proposals in DM (e.g. Marantz 1991/2000; Embick and Noyer

14We thank Judith Aissen for suggesting to us that morphological tone in possessives might be linked to genitive Case.15This observation is hinted at in Can Pixabaj (2006:64), and is implicit in some of the examples provided there. Our own

fieldwork confirms that the restriction to local person possessors does indeed hold.16The [aa] ∼ [i] ablaut seen in (28) is a fairly common, though morpheme-specific feature of Uspanteko nouns. For our

consultants, ablaut is optional when [kaa’] ‘grinding stone’ is possessed, so [ınka’] is also possible.

15

2007), we assume that Uspanteko makes use of a post-syntactic operation known as agr-insertion. That is,after the completion of all narrow syntactic operations, a head agr adjoins to F and copies its φ-features.agr insertion of this sort is assumed to be an arbitrary, language-specific property of Uspanteko morpho-syntax. The effect of agr-insertion is to create a second syntactic node with the φ-features of F (andindirectly, the φ-features of the possessor). We propose that this agr node is the locus of the morphologicalrealization of the ergative possessive prefixes in Uspanteko.

(30) Syntactic structure for [ın-tz’i’] ‘my dog’

FP

(DPposs: [+part])

NP

[tz’i’]

F

F: [+part]

H

Agr: [+part]

[in-]

Tone (on F) and ergative morphology (on agr) then combine with the possessed noun to form a complexmorphological word.

One might ask what role agr-insertion plays in the larger morpho-syntax of Uspanteko. Our basicintuition is that the presence of an ergative prefix simply indicates that a spec-head agreement relation hastaken place. (Recall that verbal ergative prefixes index the subjects of transitive verbs, which presumablyoriginate in a specifier of VP/vP; e.g. Manzini 1983; Kitagawa 1986; Woolford 1997, and much subsequentwork.) In other words, agr nodes attach to verbal or nominal heads whenever their specifier position isfilled. While we offer no formal account of the connection between ergative morphology (as agr-insertion)and filled specifiers, it is this connection that we would pursue in a more fully developed account of Uspantekopossessive syntax.17

3.2 Distribution of tone

Having considered the ways in which tone can be introduced, we now consider its distribution. First, wecapture the generalization that tone is never final with the constraint NonFin(T, tbu), which is undomi-nated.18

(31) NonFin(T, tbu)Assign one violation for every tone on a final TBU in the output.

17Note that, on our account, both tonal insertion and ergative morphology result from the spec-head relation holding betweenF and the possessor in [spec, FP]. One might then object to this (apparent) functional redundancy: both tone and ergativeagreement serve to ‘signal’ possession (as a spec-head relation). On the other hand, functional redundancy of this sort is oftenfound in natural language (e.g. Hockett 1966), so we find it unsurprising that possession is sometimes marked in multiple waysin Uspanteko.

18Campbell (1977) proposes the following stress placement rule for Uspanteko:

(i) V → [+stress] / C0VC0#

This rule is an important conceptual precursor to NonFin(T, tbu), in that it marks the penultimate mora (‘V’, in Campbell’snotation) as a privileged position for the realization of accent. However, this rule has several shortcomings: the default positionof stress in [CV.CV] words is final, not penultimate; the rule conflates stress and tone, which, as we have argued, should bedecoupled (though the two are interdependent); and it is standardly assumed that the syllable, not the mora, is the unit towhich stress is assigned (e.g. Hayes 1995; though see Cairns 2002 for a dissenting view).

16

(32) / in-kar, H / NonFin(T, tbu)

a. ☞ ın.kar

b. in.kar *!

[ın-kar] ‘my fish’ (Can Pixabaj 2006:64)

It is important to note that we do not want to achieve this result in terms of general extrametricality, becausedefault stress is word-final in Uspanteko.

(33) a. [alq’oom], *[alq’oom] ‘thief’b. [mewa], *[mewa] ‘fast’ (Can Pixabaj 2006:14)

It is far from clear why extrametricality would be disregarded in non-tonal words, but respected in wordsbearing tone. General extrametricality, then, cannot be the driving force behind the avoidance of final tone.(See also Green and Kenstowicz 1995 on problems with the context-specific revocation of extrametricality.)

Since tone never surfaces on monosyllabic words containing short vowels, NonFin(T, tbu) must domi-nate Max(T), the constraint banning tone deletion.19

(34) No tone on CV(C) words: NonFin(T, tbu) ≫ Max(T)

/ CVC, H / NonFin(T, tbu) Max(T)

a. ☞ CVC *

b. CVC *!

In Section 2.1 we saw that default final stress placement can be violated in the presence of tone, becausestress and H tone must coincide. We capture this fact with the constraint *Unstressed-H (de Lacy 2002),which states that H tone must appear in a stressed syllable (see also Hayes 1995:279).

(35) *Unstressed-HAssign one violation for every H tone on an unstressed syllable in the output.

19Without further elaboration, the ranking NonFin(T, tbu) ≫ Max(T) predicts that CV(C) roots could have ‘latent’ tone:in the isolation form of the root, underlying tone would be deleted rather than appear on the final mora; but the addition ofan affix would allow stem-final tone to surface by insulating it from word-final position.

(i) Latent tone on nonce forms [pok] and [pok-a-j]

a. / pok / → [pok]b. / pok-a-j / → [pokaj]

To the best of our knowledge, there are no words in Uspanteko that manifest latent tone in this way. While our analysis doespredict that latent tone should be possible, we believe that the lack of underlying /CV(C), H/ roots is essentially an accidentalgap. Most root-types (e.g. verb roots, positional roots, etc.) cannot appear in their unaffixed, isolation forms to begin with.While nominal roots can appear in isolation, there are very few productive nominal affixes, and many nominal affixes bear toneindependently, thus obscuring any trace of latent tone on the noun itself.

For example, plural [-ib’] and instrumental [-b’Vl] trigger tone (Can Pixabaj 2006:60), as do the local person ergativepossessive prefixes (Section 3.1). The third-person ergative prefixes [j-] and [r-] do not trigger tone; but since they do not adda TBU either, they would never cause latent tone to appear. The semi-productive abstractivizing suffix [-VVl] does not triggertone on its own, but nouns bearing [-VVl] are obligatorily possessed, and may thus bear tone for other reasons (Can Pixabaj2006:130). Finally, while the verbalizing suffixes that apply to nouns are toneless (e.g. the [-(a)aj] suffix of derived transitives,Can Pixabaj 2006:123), they are also not fully productive.

If Uspanteko ever had CV(C) noun roots with latent tone, it seems plausible that the toneless, bare forms would havebeen more frequent than the affixed forms; and further that any tone in affixed nouns could often be attributed to the affixitself. Over time, then, any words with latent tone may have been reanalyzed as simply toneless, in accord with their isolationforms. We thank Larry Hyman for bringing the question of latent tone to our attention.

17

(36) / in-kar, H / NonFin(T, tbu) *Unstressed-H

a. ☞ ın.kar

b. ın.kar *!

c. in.kar *!

So far we’ve seen that stress will shift one syllable leftward in order to coincide with tone on a penultimatesyllable. We should ask, then, why stress never shifts two syllables leftward, satisfying *Unstressed-H andlicensing tone on an antepenultimate syllable.

(37) a. [in-chıkich] ‘my large basket’b. *[ın-chikich] (Can Pixabaj 2006:61)

(38) a. [ajwu] ‘proprietor’b. /ajaw + ub/ → [ajaw-ub] ‘heads of a cofradıa’c. *[ajaw-ub] (Can Pixabaj 2006:66)

(39) a. [lekej] ‘up’b. /leke + l + ik/ → [lekelik] ‘to be high up’c. *[lekelik] (Can Pixabaj 2006:22,307)

Example (37) shows that tone appears on the penult even when the antepenultimate syllable is the possessiveprefix associated with the appearance of tone in the first place. For examples (38) and (39), shifting stressto the antepenult would allow the derived forms *[ajaw-ub] and *[lekelik] to preserve the tone placementfound in the stem forms [ajwu] and [lekej]—presumably a desirable result from the perspective of paradigmuniformity, lexical access, etc. (e.g. Steriade 2000). Since NonFin(T, tbu) will be satisfied in either case,the restriction to final or penultimate tone remains unexplained.

One possibility that we can immediately discard is that there is a high-ranked constraint directly aligningtone to the right edges of words, say Align-R(H, ω). For Uspanteko, positing such a constraint would missan important generalization: tone coincides with stress, and stress is drawn to the right edge of the wordeven in the absence of tone. An explanatory analysis of Uspanteko phonology must therefore account for thefact that tone placement is derivative of the pressures that independently govern stress placement.20

The core intuition of this analysis is that stress placement, tone placement, and stress shift in Uspan-teko all emerge from strict constraints on the realization of foot structure. It is the interaction of thesemetrical and tonal constraints that derives the two-syllable accent window of Uspanteko. To begin, weassume that footing is iambic, non-iterative, and right-aligned in Uspanteko.

(40) Iambic footing

a. [(chen.kleen)] ‘lame’b. [i(ti.neb’)] ‘place for bathing’c. [xri.xo(qi.laaj)] ‘he made her his wife’d. [(o.keb’)] ‘entry’ (Can Pixabaj 2006:21–2,52,124)

The assumption of iambic footing and right-alignment is motivated by the fact that default stress is word-finalin Uspanteko, whether the final syllable is light CV(C) or heavy CVV(C). A trochaic analysis of Uspantekostress wrongly predicts that words ending in two light syllables should have penultimate stress. This isincorrect: words ending in two [CV] syllables have default final stress.

(41) a. [la(jo.ri)] ‘today’b. *[la(jo.ri)]

A further piece of evidence for iambic footing comes from the distribution of long vowels. In Uspanteko longvowels may only appear word-finally (Can Pixabaj 2006:46; see also the appendix). As a consequence, the

20Furthermore, if violations of alignment constraints are reckoned categorically (McCarthy 2003b), Align-R(H, ω) would notby itself guarantee that stress shifts at most one syllable to the right.

18

only logically possible right-edge feet are of the shape (L L), (L H), (L), or (H) (where L = light, monomoraicsyllable, and H = heavy, bimoraic syllable). This set of foot shapes is of course suspiciously reminiscent ofthe crosslinguistic inventory of well-formed quantity-sensitive iambs (Prince 1991; Kager 1993a; Hayes 1995).Finally, non-iterativity is justified by the fact that Uspanteko lacks secondary stress, as can be seen in wordsof three or more syllables (e.g. (40-c) above).

In OT terms, both right-alignment and non-iterativity fall out from the assumption that the constraintAll-Ft-R (Prince and Smolensky 1993/2004; McCarthy and Prince 1993) is undominated.21

(42) All-Ft-R undominated in Uspanteko

/ ti-chomoor-sa-j / All-Ft-R

a. ☞ ti.cho(mor.saj)

b. ti(cho.mor)saj *!

c. (ti.cho)(mor.saj) *!

[tichomorsaj] ‘they are thinking’ (Can Pixabaj 2006:606)

Assuming further that Iamb, the constraint enforcing right-headed feet, outranks its mirror-imageTrochee,we straightforwardly derive the default stress system of Uspanteko.

(43) Default iambic stress in Uspanteko: Iamb ≫ Trochee

/ ti-chomoor-sa-j / All-Ft-R Iamb Trochee

a. ☞ ti.cho(mor.saj) *

b. ti.cho(mor.saj) *!

c. ti(cho.mor)saj *! *

d. (ti.cho)(mor.saj) *! **

Default final stress obtains in two situations: words which lack tone, and words which bear tone on a finallong vowel. In the absence of tone, tone-stress constraints like *Unstressed-H can exert no effect on themetrical structure of a word, so default stress placement results.

(44) No tone: default stress

a. [(o.kox)] ‘mushroom’b. [(chu.kej)] ‘cramp’ (Can Pixabaj 2006:24-5)

(45) No tone: *Unstressed-H inactive

/ okox / All-Ft-R Iamb *Unstressed-H

a. ☞ (o.kox)

b. (o.kox) *!

When tone appears on a final long vowel, tone already coincides with the position of default stress, so finalstress is once again unperturbed. For words ending in a long vowel, penultimate stress always violates bothIamb and Weight-to-Stress (WSP; Hayes 1981; Prince 1991; Prince and Smolensky 1993/2004), and istherefore impossible.

(46) Tone on long vowel: default final stress

21If one were inclined to be more agnostic about iterative footing in Uspanteko, the constraint All-Ft-R could be replacedwith a constraint like Align-Head-R, which demands final primary stress but allows for non-final feet (McCarthy and Prince1993; Pater 2000). Tone-triggered penultimate stress would then require the ranking {NonFin(T, tbu),*Unstressed-H} ≫Align-Head-R. At present there is no empirical evidence for non-head feet in Uspanteko, so we believe that the burden of proofis on those who assume iterative footing.

19

a. [(in.wuuj)] ‘my paper’b. [(in.kuuk’)] ‘my squirrel’ (Can Pixabaj 2006:69)

(47) No tone-stress interaction for long vowels

/ in-siip, H / NonFin(T, tbu) WSP Iamb *Unstressed-H

a. ☞ (in.siip)

b. (in.siip) *! *!

c. (in.siip) *!

[in-siip] ‘my tick’ (Can Pixabaj 2006:69)

Even without assuming a ranking between Iamb, WSP, and *Unstressed-H, these three constraints con-spire to ensure that tone never falls on the penult if the final syllable contains a long vowel. As shown in (47),such candidates must violate either *Unstressed-H or Iamb/WSP, and are therefore correctly eliminated.We can already see how the interaction of tone-stress constraints and constraints on foot structure derivesthe right-edge orientation of tone placement in Uspanteko.

The only deviation from default final stress occurs when tone appears on a word with a short vowel inthe final syllable. In just those cases tone appears on the penult, and stress retracts one syllable to alignwith tone.

(48) Tone on short vowel: stress and tone retract to penult

a. [lekej] ‘up’b. [ak’el] ‘child’ (Can Pixabaj 2006:59)

We propose that tone-driven stress retraction in Uspanteko is in fact a tone-driven iambic-trochaic reversal:to avoid placing tone on a final vocalic mora, Uspanteko reverses the headedness of final feet, yieldingpenultimate stress and tone.22

20

(49) Tone on short vowel: iambic-trochaic reversal

a. [(le.kej)] ‘up’b. [(a.k’el)] ‘child’

Formally, the constraint penalizing final tone (NonFin(T, tbu)) and the constraint forcing tone and stressto coincide (*Unstressed-H) must both outrank Iamb, which prefers right-headed feet.

(50) Tone on penult drives iambic-trochaic reversal:{Max(T), *Unstressed-H, NonFin} ≫ Iamb

/ lekej, H / Max(T) *Unstr-H NonFin(T, tbu) Iamb

a. ☞ (le.kej) *

b. (le.kej) *!

c. (le.kej) *!

d. (le.kej) *!

As (50) shows, Max(T) must also dominate Iamb to ensure that underlying pitch accent surfaces even whenrealizing tone requires the construction of trochaic feet.

To summarize, we are claiming (i) that penultimate tone results from constraints barring tone on a word-final vocalic mora; (ii) that deviations from default stress occur in order to align stress with non-final tone;and (iii) that stress retraction in Uspanteko involves an iambic-trochaic reversal; that is, stress retraction isfoot-bounded. We are now in a position to explain why stress and tone never appear farther to the left thanthe penult. Antepenultimate accent would either require tone to appear outside of a foot, and thus on anunstressed syllable; or it would require leftward shift of the default right-aligned foot. The first alternativeis ruled out by *Unstressed-H, and the second by All-Ft-R.

(51) No antepenultimate accent:{All-Ft-R, Max(T), *Unstressed-H} ≫ {Iamb, Ident(T)}

/ leke-l-ik / All-Ft-R Max(T) *Unstr-H Iamb Ident(T)

a. ☞ le(ke.lik) * *

b. le(ke.lik) *!

c. le(ke.lik) *!

d. (le.ke)lik *! *

e. (le.ke)lik *! *

[lekelik] ‘to be high up’, from [lekej] - ‘high up’ (Can Pixabaj 2006:58, 124)

(52) Ident(T)(McCarthy and Prince 1995)Assign one violation for every input-output pair {Ti, To}, such that Ti and To are tones standingin a correspondence relation and are associated with different tone-bearing units.

The complete absence of antepenultimate accent is thus a direct consequence of the pressure for right-alignedfeet, and the pressure for stress and tone to coincide.23

As suggested above, the distribution of long vowels in Uspanteko provides further evidence that accent issubject to austere constraints on foot structure. Long vowels only appear word-finally in Uspanteko, and are

22Foot-form reversals of this sort—sometimes known as ‘rhythmic reversals’—have also been proposed for Choctaw, SouthernPaiute, Ulwa, Axininca Campa (Prince and Smolensky 1993/2004:58), Tiriyo Carib (van de Vijver 1998:Ch.2), Hopi (Gouskova2003:Ch.3), Nanti (Crowhurst and Michael 2005), Panoan languages (Elias-Ulloa 2006), Takia (de Lacy 2007), and Awajun(McCarthy 2008).

23Rather than assume right-aligned feet, one might entertain a foot-free analysis of this two-syllable accent window byappealing to a pressure to avoid word-final lapses (e.g. Kager 2001, 2005). However, see Section 3.3 for segmental evidence thatUspanteko accent does indeed depend on metrical foot structure.

21

therefore always stressed. We take this fact as evidence that the constraint Weight-to-Stress is active inUspanteko: long vowels must be stressed. Since stress retraction is foot-bounded, the only long vowels thatcould potentially bear stress are those long vowels appearing in the last two syllables of the word. Finalstress is of course perfectly licit; but placing stress on a penultimate long vowel would require a violation ofIamb. As tableau (53) demonstrates, the ranking Iamb ≫ Id(Length) guarantees that shortening non-finallong vowels will be preferred to retracting stress to a long vowel in the penult.

(53) No non-final long vowels: {All-Ft-R, WSP, Iamb} ≫ Id(Length)

/ x-r-elk’waal-aj / All-Ft-R WSP Iamb Id(length)

a. ☞ xrel(k’wa.laj) *

b. xrel(k’waa.laj) *!

c. xrel(k’waa.laj) *!

d. (xrel.k’waa)laj *!

[xrelk’walaj] ‘I sired him/her’, from [alk’waal] ‘son’(Can Pixabaj 2006:123)

The only way a non-final long vowel could be prosodified would be by violating All-Ft-R, WSP, or Iamb,all of which dominate Id(Length). Non-final long vowels are thus repaired via vowel shortening, leavingdefault stress assignment intact. Here we see an important difference in the relative prominence of toneand vowel length: iambic reversals can be conditioned by tone (high-ranked NonFin(T, tbu)), but not bylength (low-ranked Id(Length)).

Given this dichotomy, one potentially problematic form would be one in which a non-final long vowelwere allowed to surface unaltered by virtue of bearing tone.

(54) Tone + length 6= non-final long vowel

/ juun-kitz / NonFin(T, tbu) WSP Iamb Id(length)

a. / (jun.kitz) * *!

b. ☞ (juun.kitz) *

[junkitz] ‘a little’ (Can Pixabaj 2006:58, 144)

In (54) a non-final long vowel ‘piggybacks’ on the iambic-trochaic reversal driven by non-final tone, and isthus indirectly licensed. Note, however, that the troublesome form *[juunkitz] contains an (H L) trochee.As argued in Hayes (1981, 1995); Prince (1991); Kager (1993a,b, 1999), and Mester (1994), (H L) trocheesare marked relative to the bimoraic even trochee (L L). One way to rule out this dark horse candidate, then,is to assume that the constraint *UnevenTrochee is active in Uspanteko. (See Pruitt 2010:505 for a briefoverview of proposals for capturing the preference for balanced/even trochees in OT.)

(55) *UnevenTrocheeAssign one violation for every trochaic foot of the shape (HL), where H = heavy (bimoraic) syllableand L = light (monomoraic) syllable.

(56) Tone can’t save non-final long vowels: *UnevenTrochee ≫ Id(Length)

/ juun-kitz / Iamb *UnevenTrochee Id(length)

a. ☞ (jun.kitz) * *

b. (juun.kitz) * *!

Iambic-trochaic reversals in Uspanteko thus instantiate an interesting ‘Emergence of the Unmarked’ effect(McCarthy and Prince 1994): reversals are permitted if and only if they result in the least-marked trochaicform (L L). Here again we find that Uspanteko imposes strict and varied requirements on metrical structure,

22

despite the descriptive simplicity of the accentual system.This section provided arguments for the following constraint rankings, which are graphically represented

with a Hasse diagram in Figure 10:

(57) a. NonFin(T, tbu) ≫ Max(T) (34)No tone on monosyllabic CV(C) words containing short vowels.

b. Iamb ≫ Trochee (43)Iambic footing.

c. NonFin(T, tbu) ≫ Ident(T) (not shown)Prevents tone from surfacing faithfully on word-final moras when underlyingly specified in poly-syllabic words.

d. Max(T), *Unstressed-H, NonFin(T, tbu) ≫ Iamb (50)Tone-driven iambic-trochaic reversals.

e. All-Ft-R ≫ Iamb (51)No tone-driven leftward foot displacement.

f. All-Ft-R, Max(T), *Unstressed-H ≫ Ident(T) (51)Underlying pre-penultimate tone surfaces on the stressed penult/ultima.

23

g. All-Ft-R, WSP ≫ Id(Length) (53)24

No non-final long vowels.

h. *UnevenTrochee ≫ Id(Length) (56)Penultimate long vowels always shorten, even when accented.

*UnevenTrochee WSP All-Ft-R *Unstressed-H NonFin(T, tbu)

Max(T)

Id(Length) Iamb Ident(T)

Trochee

Figure 10: Hasse diagram of ranking arguments

Taking a broader perspective, in this section we have demonstrated that stress placement, tone placement,and the distribution of long vowels in Uspanteko are all determined by simple, crosslinguistically plausibleconstraints on metrical structure. It is a remarkable fact that Uspanteko is sensitive to foot-based metricalrestrictions of exactly the same type as found in languages with greater surface complexity in their accentualsystems. We defend this claim in the following section, where we provide further segmental evidence for footstructure in Uspanteko. In section Section 4 we address additional tone-segment interactions, focusing oncases where segmental structure interferes with the realization of tone.

3.3 Segmental evidence for foot structure

3.3.1 Tone-driven reversals and perfect prosodic word effects

Section Section 2.2 established the generalization that tone in bisyllabic words is robustly correlated withtheir sonority profile (Can Pixabaj 2006:58).25 In particular, bisyllabic words tend to have tone under anyof the following conditions: (i) the final syllable, if unstressed, would have a low sonority head, normally[i] or reduced 〈a〉/[@]; (ii) the penult, if stressed, would have a high sonority [a] head; or (iii) the sonorityprofile across the word is even, in the sense that the penult and final syllable are headed by vowels of equalsonority. In this section we analyze the interaction between tone and vowel sonority in bisyllabic words, andargue that such interactions are mediated by foot structure. The core idea is that the ideal metrical shapefor phonological words in Uspanteko is a single bimoraic foot with a sonority profile that does not rise intothe weak branch. Tone insertion in bisyllabic roots is licensed when it produces prosodic words meetingthese conditions. We start with the sonority generalizations.26

We assume that vowel sonority is a function of height, where low vowels are more sonorous than mid vow-els, which are in turn more sonorous than high vowels and schwa (e.g. Jespersen 1904; Dell and Elmedlaoui1985; Clements 1990; Prince and Smolensky 1993/2004, etc.).

(58) Relative sonority scale for vowels:Low > Mid > High/@

24The ranking *UnevenTrochee ≫ Ident(Length) undermines the argument for Iamb ≫ Ident(Length) in (53). Conse-quently, the latter ranking is omitted here.

25Throughout this section, ‘bisyllabic’ refers only to forms consisting of two light syllables, i.e. words with a short vowel inthe final syllable.

26It is important to note that, according to Can Pixabaj 2006, all of these generalizations are statistical. In combing throughthis work and the Uspanteko dictionary (Mendez 2007), we have only been able to find one clear counterexample. Ideally wewould be able to provide a statistical analysis of this area of the lexicon, but large balanced corpora of Uspanteko that reliablyindicate tone do not exist. In the future we hope to be able to quantify the strength of these generalizations. In the meantime,we feel that the fact that native speakers find it easy to think of examples obeying these generalizations, but hard to think ofcounterexamples, is enough to motivate an analysis of these data.

24

In Uspanteko, tone appears on the vast majority of bisyllabic words with a low sonority vowel nucleus ([i]or 〈a〉/[@]) in their final syllable. We transcribe unstressed /a/ as [@] in the following examples to highlightthis generalization.

(59) a. [anim] ‘woman’b. [r-uxib’] ‘his/her/its aroma’c. [ısim] ‘stamp’

d. [saq’@j] ‘summer’e. [tun@q’] ‘Adam’s apple’f. [ıj@j] ‘seed’ (Can Pixabaj 2006:58)

If vowel sonority truly conditions the presence of tone in such words, we should expect final [u] to patternwith [i] and [@] in triggering tone, given that [u] is also a low-sonority high vowel. It is difficult to findrelevant examples, but at least one word suggests that this prediction is correct.

(60) [ınup] ‘ceiba (species of tree)’ (Mendez 2007)

In addition, bisyllabic words with the highest sonority vowel, stressed [a], as the initial syllable nucleusoverwhelmingly bear tone. This pattern holds regardless of the sonority of the final syllable nucleus.

(61) a. [ab’@j] ‘stone’b. [pat@n] ‘burden’

c. [ajchi’] ‘matazano (species of tree)’d. [balik] ‘brother-in-law (for a man)’

e. [ak’el] ‘child’f. [chakej] ‘dry’

g. [awus] ‘fava bean’h. [ajwu] ‘owner’ (Can Pixabaj 2006:58–59)

Finally, tone also appears if both syllable nuclei in a bisyllabic word are of equal sonority.

(62) a. [lekej] ‘up’b. [tz’unun] ‘hummingbird’c. [ojor] ‘a long time ago’ (Can Pixabaj 2006:59)

The core generalization here is that, with overwhelming frequency, bisyllabic roots with falling or levelsonority between successive vowels also bear tone. What we never find is a bisyllabic word with tone and arise in vowel sonority from the tonic to post-tonic syllable (i.e. a penult high vowel with tone followed by afinal mid vowel). For example, roots like the following are always toneless.27

(63) a. [ixk’eq] ‘nails’b. [ikeq’] ‘twine sling’

c. [chukej] ‘cramp’d. [uke] ‘guachipilın (species of plant)’

Note that words like [ixk’eq] are perfectly capable of bearing tone: derived forms like [w-ıxk’eq] ‘my nails’show that the sonority requirements holding of unaffixed roots do not apply to morphologically complex

27We have been able to find a few counterexamples, but most are weak at best. For example, the word [sıner] ‘dinner’, whichis a borrowing from the Spanish cena, should not have tone according to our generalizations; however, this example does notpose a real problem for our account because Spanish penultimate stress is always reinterpreted as tone in borrowings. Similarly,there are words that have been reported inconsistently in the grammatical literature, like ‘his leg’, which is found written bothas [raqan] and as [raqan]. The only firm counterexample we have found is [chukuy] ‘pine fruit’, which does not bear tone despitehaving two identical short vowels.

25

forms. It seems that the tone-sonority interactions described above are limited to bare roots alone.Stepping back, what we see in these examples is more evidence for the foot-based analysis of Uspanteko

tone. If we try to conceive of the least marked trochaic foot, it would be a bisyllabic LL foot with everypossible prominence contrast favoring its left branch. This is exactly what these distinguished words areconverging on in Uspanteko. They are bisyllabic LL forms where the left syllable bears both tone and stress,and is no lower in sonority than the right syllable. If we want a phonological account of the presence of tonein these examples, the analysis needs to insert tone just in bisyllabic words without a sonority rise across thefoot. While the most important conclusion of this section lies is the generalization itself, namely assumingthat tone insertion entails an iambic-trochaic reversal makes of these data, we propose that this accentualpattern reflects a constraint preferring ‘perfect prosodic words’ (Zec 1999; Ito and Mester 2011).

As discussed in more detail below, Perfect Prosodic Word (hereafter PPW) is a markedness con-straint assigning special status to words that are coextensive with a single foot, and that also meet additionaldemands on the relative prominence of syllables within that foot. We will present the PPW account first,and then show that an alternative account that decomposes PPW into its constituent constraints runs intoproblems.

Consider the first feature of PPW, the requirement that each prosodic word ω correspond to exactly onefoot. As observed by Zec (1999) (and discussed in Ito and Mester 2011), suffix-triggered vowel shorteningin the Neo-Stokavian dialect of Serbo-Croatian occurs if and only if the resulting complex corresponds to abimoraic trochee. (This pattern of shortening is apparently specific to the adjective-forming suffixes [-ost]and [-ask].)

(64) Suffix-triggered shortening and PPW effects in Serbo-Croatian (Zec 1999)

a. [mlaad-] → [mladost-] ‘young’b. *[mlaadost-]c. [ziiv-] → [zivost-] ‘lively’d. *[ziivost-]

e. [humaan-] → [humaanost-] ‘humane’f. *[humanost-]g. [opaak-] → [opaakost-] ‘vicious’h. *[opakost-]

As the examples in (64) illustrate, vowel shortening applies if the result is a bimoraic trochee like [(mla.dost)],but not if the resulting form would be larger than a bimoraic trochee, as in *[o(pa.kost)]. Zec captures thisfact with a constraint requiring certain morphological domains to correspond exactly to a single foot (thus*[(mlaa)dost]); following Ito and Mester (2011), we interpret this phenomenon as an expression of PPW.28

Ito and Mester (2011) also show that the distribution of Danish stød (essentially a phonologically-drivenpitch accent) follows from similar pressures on the alignment of prosodic words and feet.

Languages that are sensitive to PPW effects can also place restrictions on the prosodic shape of therelevant foot. Ito and Mester (2011) argue that the strong statistical preference for initial pitch accent onbimoraic words in Tokyo Japanese is also due to the workings of PPW. While accent placement is foot-based in Tokyo Japanese (e.g. Kubozono 2008), there is no general requirement that foot heads correspondto accented syllables (i.e. there is widespread covert footing). Bimoraic words are of course those words that,in principle, could satisfy the size requirement of PPW. The fact that bimoraic words bear initial accent,however, must be attributed to an independent pressure for the foot-head to be phonetically salient in PPWcontexts (footing is trochaic in Tokyo Japanese). The intuition at work here is that prosodic words are only‘perfect’ if they correspond to a single foot, and the head of that foot is phonetically prominent (cf. Zec’s1999 FootSalience constraint; see Teeple 2009 for closely related ideas).

We claim that the perfect prosodic word in Uspanteko is coextensive with a bisyllabic foot that has anon-rising sonority profile and a head that bears tone. We define the constraint PPW as in (65), whichdiffers slightly from the formulation proposed by Ito and Mester (2011).

28It bears mentioning that these shortening effects cannot be attributed to the constraint *UnevenTrochee (55), since*UnevenTrochee would wrongly favor shortening in [o(paa.kost)], *[o(pa.kost)] as well.

26

(65) Perfect Prosodic Word (PPW)Assign one violation mark for every prosodic word ω that does not meet all of the following criteria:

(i) ω is coextensive with a single foot F.(ii) The head syllable of F (σS) bears tone.(iii) F is bisyllabic.(iv) The nucleus of σS is at least as sonorous as the nucleus of σW , the syllable occupying the

weak branch of foot F.

Clause (i) captures the basic size requirement behind PPW effects (more on this below). It is importantto recognize that clause (i) is not a general condition forcing prosodic words to be of some minimal size.Trisyllabic words, for example, necessarily violate clause (i), because the foot in a [σ(σσ)] structure is notcoextensive with the entire prosodic word.

Clauses (ii)–(iv) express the second facet of PPW, namely the conditions it places on the prosodic shapeof the single foot in the prosodic word. Our general claim is that PPW forces foot heads to have greaterphonetic prominence than foot non-heads in PPW contexts. Uspanteko expresses relative prominence by acombination of tone and restrictions on vowel sonority; in other languages, PPW might be satisfied by othermeans, as made available by the phonetics and phonology of the language. Japanese, for instance, uses pitchto satisfy the relative prominence clause of PPW, but opts not to make use of vowel sonority in the sameway.29

What of clause (iii), the restriction to bisyllabic feet in PPW contexts? This clause captures the factthat monosyllabic CVV(C) roots—which are necessarily parsed as a single foot—do not all bear tone. SinceCVV(C) roots satisfy both clause (i) and clause (iv) of PPW (the latter vacuously), clause (iii) is needed toguarantee that tone does not appear on such roots with the same regularity as for bisyllabic roots.30 Butwhy might something like clause (iii) hold? Here we appeal to the long-standing idea that prominence is arelational notion (e.g. Liberman 1975): no phonological element is ‘prominent’ in an absolute sense, onlymore or less prominent than other elements within the same phonological structure. Just as tone insertionand conditions on relative vowel sonority serve to highlight foot heads in PPW contexts, we believe that thebisyllabicity requirement (iii) enhances the prominence of accented syllables by ensuring that a syllable withlow phonetic salience will appear within the same foot.31 In other words, the bisyllabicity requirement forcesan explicit comparison between the phonetically salient foot head and the less salient non-head, therebyemphasizing the prominence of the foot head itself.32

Note that we are not claiming that Uspanteko prefers trochaic footing over iambic footing in PPWcontexts (which would be at odds with the general preference for iambs in the language). Trochaic footingemerges in PPW contexts under the interaction of constraints governing the relative prominence of footheads (i.e. foot heads should bear tone), and the non-finality constraint preventing tone from appearing ona final short vowel.33

With the basic structure of PPW in hand, we can now see how it captures the tone distributions outlined

29Uspanteko also differs from Danish and Serbo-Croatian in that PPW effects are limited to monomorphemic roots. Forexample, complex forms like [san-s-ik] ‘swollen’ [k’iy-naq] ‘grown’ do not bear tone, while simplex forms like [ab’aj] ‘stone’ do(Can Pixabaj 2006:97,156–7).

30See Section 4.2 for evidence that tone is also dispreferred on final long vowels for independent reasons. In connection withthis fact, note that bisyllabic roots containing a final long vowel (e.g. [tu.kuur] ‘owl’, Can Pixabaj 2006:38) do not bear tonewith any notable frequency, even though such words can in principle satisfy all four clauses of PPW (65). We assume that the

relative markedness of tonal [. . . CVVC#] syllables in Uspanteko masks PPW effects in such words.31Along these lines, Kenstowicz (1994); Gouskova (2003); Zec (2003) and de Lacy (2004, 2007) (among others) have suggested

that feet may impose different sonority requirements on their strong and weak branches, with a clear preference for high-sonorityheads and low-sonority non-heads. Teeple (2009) argues at length that prominence constraints within a phonological domain(like the foot) should refer to both prominent and non-prominent positions simultaneously.

32The bisyllabicity requirement also appears to be unique to Uspanteko: in Japanese, for example, PPW makes no distinctionbetween monosyllabic and bisyllabic two-mora words (Ito and Mester 2011). It may be relevant that pitch accent in TokyoJapanese is an HL contour, and in a certain sense has relative prominence ‘built in’ to the accent itself. See also footnote 30for an alternative explanation of the bisyllabicity requirement.

33Curiously, apart from Uspanteko PPW effects have so far only been observed for languages with trochaic footing. Moreresearch is needed to determine whether this is a real generalization about the content of PPW, or an artifact of limiteddata. Also interesting in this regard is the fact that “. . . trochaic systems tend to be characterized by alternations in pitchand intensity, while iambic systems are marked by alternations in length” (Goad and Buckley 2006:115, citing Hayes 1995). InPPW contexts in Uspanteko, we find trochaic feet that show prominence asymmetries in both pitch and intensity (≈ sonority),in line with this general finding about the expression of prominence in trochaic stress accent systems.

27

above. PPW forces tone to appear on the foot head; since high-ranked NonFin(T, tbu) prevents tonefrom appearing on the final syllable, a rhythmic reversal occurs to accommodate tone in the usual way.

(66) PPW effects: {NonFin(T, tbu), PPW} ≫ {Iamb, Dep(T)}

/ anim / NonFin(T, tbu) PPW Iamb Dep(T)

a. ☞ (a.nim) * *

b. (a.nim) *! *

c. (@.nım) *! *

d. (a.nim) *!

[anim] ‘woman’ (Can Pixabaj 2006:58)

If any of clauses (i)–(iv) cannot be satisfied—say, if a bisyllabic word has the wrong sonority profile when itbears tone, necessarily violating either clause (ii) or clause (iv)—then tone fails to appear.

(67) Tone blocked by conditions on vowel sonority: NonFin(T, tbu) ≫ PPW

/ ikeq’ / NonFin(T, tbu) PPW Iamb Dep(T)

a. ☞ (i.keq’) *

b. (i.keq’) *! *

c. (ı.keq’) * *! *!

d. (i.keq’) * *!

[ikeq’] ‘twine sling’ (Can Pixabaj 2006:55)

To close, we return to clause (i) of PPW, the requirement that a prosodic word correspond exactly to asingle foot. To demonstrate that this clause is in effect, we would need to show that monomorphemic wordslarger than a single foot (e.g. trisyllabic roots) do not have any special tendency to bear tone. There arenot many such words because Mayan languages favor monosyllabic and bisyllabic roots. Examples do exist,though, and they do not have tone, even with the correct sonority profile.

(68) a. [ixpaqar] ‘toad’b. [ixnakar] ‘wild onion’c. [lajori] ‘today’d. [aware] (surname)e. [chamatun] (surname)

Even though the final two syllables constitute a foot with the right sort of sonority profile, these words do notbear tone. Our account makes the right prediction because inserting tone or altering default stress cannotprevent these words from violating PPW. They are simply too long.

(69) / lajori / NonFin(T, tbu) PPW Iamb Dep(T)

a. ☞ l@(jo.ri) *

b. l@(jo.ri) * *!

c. l@(jo.ri) * *! *

d. l@(jo.rı) *! * *

[lajori] ‘today’ (Can Pixabaj 2006:163)

While the PPW account has crosslinguistic support and successfully captures the patterns we see inUspanteko, one might challenge it on methodological grounds. Since OT derives its predictive power fromconstraint interaction, we should prefer an account with many separate constraints over an equivalent accountthat has one constraint with many clauses. What we will show now is that a decompositional account of the

28

Uspanteko facts must resort to stipulative constraint indexing, essentially recapitulating the PPW account.The most obvious way to decompose PPW is to assume that that there are active constraints penalizing

toneless foot-heads and foot-heads that are less sonorous than foot-non-heads. Since we build only onerightmost foot in Uspanteko, when its right branch is more sonorous than its left branch, we get iambicstress and no tone because of high ranking NonFin(T, tbu). When its left branch is equally sonorousor more sonorous than its left branch, we get an iambic-trochaic reversal that allows the foot head to beboth tone-bearing and relatively sonorous within the foot. To implement this analysis we need the followingconstraints penalizing foot-heads deficient along some relative prominence requirement. We borrow theconstraint format in Teeple (2009), though the discussion is not a commentary on the proposals within thatwork.

(70) PROM(σ, Ft) ⇔ Tone-bearing (Prom/T)Assign one violation mark for every toneless σS .

(71) PROM(σ, Ft) ⇔ Sonorous (Prom/Son)Assign one violation mark for every σS with a nucleus less sonorous than the nucleus of σW in thesame foot.

When these constraints outrank Iamb and Dep-T, we correctly predict that bisyllabic words with the rightsonority profile should bear tone.

(72) {NonFin(T, tbu), PROM/Son, PROM/T} ≫ {Iamb, Dep(T)}

/ anim / NF(T, tbu) PR/Son PR/T Iamb Dep(T)

a. ☞ (a.nim) * *

b. (a.nim) *! *

c. (@.nım) *! *

d. (a.nim) *!

[anim] ‘woman’ (Can Pixabaj 2006:55)

The fact that tone is not always realized provides a ranking argument that Prom/Son outranks Prom/T.When the right branch of the foot has a more sonorous nucleus, it’s better to foot an iamb than to realizetone on a trochee with a bad sonority profile.

(73) {NonFin(T, tbu), PROM/Son} ≫ PROM/T ≫ {Iamb, Dep(T)}

/ ikeq / NF(T, tbu) PR/Son PR/T Iamb Dep(T)

a. ☞ (i.keq) *

b. (ı.keq) *! * *

c. (i.keq) *! * *

d. (i.keq) *! *

[ikeq] ‘twine’

This ranking further predicts that when both vowel nuclei are of equal sonority, it is best to foot a trocheebearing tone. The reason is that PROM/Son will prefer neither trochaic nor iambic footing in such cases.

(74) {NonFin(T, tbu), PROM/Son} ≫ PROM/T ≫ {Iamb, Dep(T)}

29

/ ojor / NF(T, tbu) PR/Son PR/T Iamb Dep(T)

a. ☞ (o.jor) * *

b. (o.jor) *!

c. (o.jor) *! *

d. (o.jor) *! *

[ojor] ‘a long time ago’ (Can Pixabaj 2006:59)

While the decompositional account makes the right predictions in bisyllabic words, when we move tolonger words the analysis makes pathological predictions with no easy solution. Consider, for instance,[lajori] ‘today’. As it stands, the decompositional analysis incorrectly predicts tone on the penultimatesyllable.

(75) / lajori / NF(T, tbu) PR/Son PR/T Iamb Dep(T)

a. ☞ l@(jo.ri) * *

b. l@(jo.ri) *! *

c. l@(jo.ri) *! *

d. l@(jo.rı) *! * *

[lajori] ‘today’ (Can Pixabaj 2006:163)

Solving this problem is very difficult, especially if we want to maintain an account in terms of relativeprominence in the foot. One approach is to exploit the fact that feet will always be initial in bisyllabicwords in Uspanteko, but not in trisyllabic words. If we parameterize PROM/Son and PROM/T so thatthey only penalize initial feet, then all trisyllabic words will be well-formed with respect to these constraints,allowing iambic footing to emerge.

(76) PROM(σ, Ftinit) ⇔ Tone-bearingAssign one violation mark for every toneless σS in a word initial foot.

(77) PROM(σ, Ftinit) ⇔ SonorousAssign one violation mark for every σS with a nucleus less sonorous than the nucleus of σW in aword initial foot.

(78) / lajori / NF(T, tbu) PR/Soninit PR/Tinit Iamb Dep(T)

a. ☞ l@(jo.ri)

b. l@(jo.ri) *! *

c. l@(jo.ri) *!

d. l@(jo.rı) *! * *

[lajori] ‘today’ (Can Pixabaj 2006:55)

This trick works, but amounts to a stipulative restatement of PPW: it happens to succeed only becauseUspanteko builds a single, right-aligned foot. There are no independent reasons to think that the heads ofinitial feet should be especially salient in Uspanteko, or any other language for that matter.34 This is in

34While there might not be pressure for initial feet to have an internal tone or sonority contrast, there is reason to believethat initial syllables favor prominent elements (e.g. Beckman 1998; Smith 2005). We could abandon the foot-based account infavor of an analysis based on edge prominence, but this would only account for the distribution of tone. That is, while it mightmake sense to have a constraint penalizing initial toneless syllables, there is no evidence for a constraint demanding that theinitial syllable be at least as sonorous as the following syllable regardless of footing. To account for the sonority facts, we wouldhave to resort to the same parameterization mechanism discussed earlier (i.e. PROM/Soninit). The result is that to build an

30

contrast to an account based on PPW effects, which are also attested in languages with iterative footing.Positional constraints like PROM/Soninit would be of little use in such cases.

Now that we’ve seen that factoring PPW into independent constraints falls short, we return to theapparently unitary nature of PPW. While we would prefer not to propose constraints with many subclauses,doing so here helps explain why three seemingly unrelated phenomena—syllable count, tone insertion, andrelative vowel sonority—are so strongly correlated in the pattern we see in Uspanteko: they are all expressionsof the independently-attested pressure for foot-heads in PPW contexts to be maximally salient.

These PPW effects also provide further support for a foot-based analysis of Uspanteko accent. A crucialfact about bisyllabic roots is that the appearance of tone depends not only on the vowel quality of the penult(where tone actually shows up), but also on the vowel quality of the final syllable. The distribution of tonein bisyllables thus has a non-local character, in that the licensing of tone on the penult is contingent onproperties of an adjacent, non-tonal syllable. This non-local dependency makes sense if couched in terms ofrelative prominence within the foot, since prosodic structure is often sensitive to domain-internal prominencerelations. The metrical foot thus reduces an apparent non-local effect to a local, domain-internal relation,and thereby captures the interaction between tone and vowel sonority in a principled way. Importantly, theargument for foot structure is independent of our arguments for a PPW-type constraint, since the relationalnature of the phenomenon implicates metrical structure whether or not one accepts our explanation for whyinteractions between tone and vowel sonority are limited to bisyllabic roots.

To summarize, the complex constraint PPW (65) is responsible for interactions between tone and vowelsonority in bisyllabic words in Uspanteko. The necessary rankings established in this section (which areconsistent with the Hasse diagram in Figure 10) are given in (79).

(79) Ranking for perfect prosodic word effects in Uspanteko:NonFin(T, tbu) ≫ PPW ≫ Iamb, Dep(T)

3.3.2 Syncope

To round out the discussion of tone and its interaction with various morphophonological, prosodic, andsegmental phenomena, in this section we present the effects of tone realization on syncope in Uspanteko.Though syncope presents a potential opacity problem for the analysis as presented, at the same time itprovides more evidence for the foot, and for the analysis of penultimate accent as a tone-triggered foot-formreversal.

For the time being we will focus on syncope in bisyllabic forms, and even then we must leave a fullaccount of syncope for future work. In fact, while we call this vowel reduction process ‘syncope’ for the sakeof concreteness, we raise the possibility that syncope of this sort preserves syllabicity, in the sense that voweldeletion may not alter abstract syllable structure when it applies (e.g. Kager 1997). What’s important hereis that the locus of vowel deletion is clearly foot-determined, and that our analysis correctly predicts whensyncope will take place in the pre-tonic or post-tonic syllable.

The generalization seems to be that syncope optionally targets the weak branch of the foot (e.g. Kager1997; Gouskova 2003; Blumenfeld 2006; McCarthy 2008), though it is constrained by the quality of the targetvowel and the resulting consonant cluster. Thus in a bisyllabic form with no tone and default stress, syncopetargets the initial syllable. This is precisely what we predict if default final stress is the result of iambicfooting.

(80) a. [simiin] ∼ [smiin] ‘ginger’b. [chukuy] ∼ [chkuy] ‘pine fruit’c. [kuwa’y] ∼ [kwa’y] ‘horse’d. [raqan] ∼ [rqan] ‘his leg’ (Can Pixabaj 2006:37)

Syncope of this sort is not simply the context-free deletion of unstressed vowels: in words with final stress,syncope only targets the immediately pretonic syllable.

(81) a. [inachape’] ∼ [inachpe’] ‘Grab me!’

account that decomposes PPW, we either have to fully replicate PPW via stipulative constraint parameterization, or partiallyreplicate it via constraint parameterization and a non-uniform analysis of the tone and sonority patterns in bisyllabic forms.

31

b. *[inchape’]c. *[nachape’]

Syncope is variable: in elicitation, speakers produce the same word both with and without vowel deletion.However, speakers still judge certain cases of syncope as ungrammatical. While we do not have enough datato fully characterize when the process can apply, syncope seems to only target low sonority vowels; that is,high vowels and 〈a〉 (which is realized as [@] when unstressed). Thus, the two mid vowels [e o] in (82) cannotbe targets for syncope.

(82) a. [keqiix] ‘dark-colored mushroom’b. *[kqiix]

c. [xinkojon] ‘I accepted it’d. *[xinkjon]

While we will not present a full analysis of the Uspanteko facts, differential syncope of this sort is common,and we could pursue a markedness-based analysis like that developed in Gouskova (2003:Ch.4).

Syncope is also blocked to avoid certain consonant clusters (see also Kager 1997). While we do not havea complete inventory of banned clusters, derived geminates are blocked, for example (McCarthy 1986).

(83) a. [jujun] ‘some’b. *[jjun]

We suggested that syncope targets the pre-tonic syllable in forms with default stress because this syllable isin the weak branch of an iamb. Further evidence that syncope takes place over default iambic footing comesfrom the behavior of bisyllabic forms with final long vowels. Instead of analyzing a form like [masaat] ‘deer’as an (LH) iamb, we could imagine an alternative analysis that foots a single (H) trochee at the right edgeof the word.

(84) a. [(ma.saat)] (iambic parse)b. [ma(saat)] (trochaic parse)

Assuming that syncope is governed by foot structure (e.g. McCarthy 2008), pre-tonic syncope in words like[masaat] would argue for iambic footing. This is exactly what we find.

(85) [masaat] ∼ [msaat] ‘deer’

The evidence from syncope thus supports our claim that default final stress is due to a right-aligned iambicfoot. Pre-tonic syllables delete under syncope because syncope targets the weak branch of the foot.

If this is the right analysis of pre-tonic syncope, we also predict that syncope should target the finalsyllable in words with penultimate tone. On our analysis, the final syllable in such words would be in theweak branch of the foot, since penultimate tone is due to an iambic-trochaic foot-form reversal. We thusexpect post -tonic syncope in words with penultimate tone, which is precisely what we find.35

(86) a. [ınchaj] ∼ [ınchj] ‘my pinetree’b. [ınpix] ∼ [ınpx] ‘my tomato’c. [ıwir] ∼ [ıwr] ‘yesterday’d. [walib’] ∼ [walb’] ‘my sister-in-law’

(87) a. [xinchakunik] ∼ [xinchakunk] ‘I worked’b. *[xinchkunik]c. *[xnchakunik]

Our account thus correctly predicts that accent shift, as a foot-form reversal, should correlate with a shiftin the locus of syncope. Analyses of tone that do not make use of the foot, or that rely on foot retractionor extrametricality to capture penultimate accent, do not predict these syncope facts. If anything, such

35Syncope in words with penultimate accent is hinted at in Can Pixabaj (2006:71), but essentially ignored.

32

approaches predict that we should always find pre-tonic syncope; that is, when tone appears on the penult,there should be syncope in the antepenultimate syllable, which is unattested in Uspanteko.

Finally, we know we want a unified analysis of pre-tonic and post-tonic syncope because they are subjectto the same segmental restrictions. For example, post-tonic syncope in tonal forms is blocked if it derives ageminate, just like syncope under default final stress.

(88) a. [ajij] ‘sugarcane’b. *[ajj]

c. [axix] ‘garlic’d. *[axx]

Similarly, we find differential syncope in tone-bearing forms. Post-tonic mid vowels resist syncope in thepresence of tone.36

36We also find ablaut feeding syncope under possession. Optional syncope yields [ınch’] for the first-person singular possessedform of [che’] ‘tree’, but it alternates with [ınchi’], not *[ınche’].

33

(89) a. [wıxkeq] ‘my fingernail’b. *[wıxkq]

c. [etzel] ‘evil’d. *[etzl]

In summary, while we do not have a complete account of Uspanteko syncope, the locus of syncope followsfrom our account, under the assumption that word-level prosody in Uspanteko always involves a right-alignedfoot. Default footing is iambic, so when syncope applies in forms with default stress it deletes the vowelnucleus of the penultimate syllable. The final vowel deletes when syncope applies to forms with penultimatetone because, as we have argued, tone insertion results in an iambic-trochaic reversal.

Of course, one complication for this view of Uspanteko syncope is that it is potentially opaque: penulti-mate accent is due to a pressure against placing high tone on a final TBU, but this is exactly the configurationthat results from post-tonic syncope. Whether there is a real opacity problem here depends on the natureof Uspanteko vowel deletion. If the phenomenon described in this section is not true syncope, but rathersyllable-preserving vowel reduction (as in Kager’s 1997 account of Macushi Carib), then there is no opacityproblem.37 If this pattern of vowel deletion is syllable-destroying syncope, then we do, in fact, have anopacity problem. The upside is that opaque syncope is a problem for which there are standing solutions.For instance, in a serialist treatment of syncope (e.g. McCarthy 2008), feet could be built at a stage deriva-tionally prior to foot-based vowel reduction and deletion. If tone placement happens while building feet,then syncope still renders tone placement opaque, but it no longer poses a formal problem for our foot-basedaccount of the facts.

In the last two sections we provided additional evidence for a foot-based account of Uspanteko accent.This evidence concerned the interaction of tone and segmental structure, especially regarding the distributionof vowels. Beyond establishing the presence of foot structure in Uspanteko, our arguments further supportedthe claim that penultimate accent in Uspanteko is the result of an iambic-trochaic ‘rhythmic reversal’.The following sections wrap up some remaining empirical issues in the accentual system of Uspanteko. Inparticular, we consider two cases where there is conflict between segmental structure and the morphologicalneed to realize tone. In some cases, tone is blocked and the root is realized faithfully; in other cases thesegmental structure of the root is altered in order to realize tone.

37The hallmark of this sort of vowel reduction is that the vowel nucleus is still active for phonological processes. While weneed to do more work to confirm whether or not syncope affects syllable-based consonant allophony (stop aspiration, nasal placeassimilation, etc.; see Can Pixabaj 2006:Ch.2), Uspanteko syncope does have a number of affinities with the syllable-preservingsyncope of Macushi Carib. For one, it is variable, and to some extent gradient: non-syncopated weak vowels are reduced tovarious degrees, and syncope seems to be an endpoint for this gradient reduction. Syncope also derives many clusters thatare otherwise unattested in the language (e.g. [chk], as in [chkuy] ‘pine fruit’). As Larry Hyman suggests to us, the fact thatsyncope derives otherwise illicit clusters (a property which it shares with schwa deletion in French; e.g. Dell 1995) might pointto the preservation of a mora or vocalic nucleus at the phonological level.

34

4 Tone-segment conflicts and nominal cophonologies

In this section we present an analysis of how the shape of a nominal root affects or is affected by the affixationof morphemes that require tone insertion. In particular we will see that there is pressure against realizingtone when the final syllable is CV’C or has a long vowel nucleus. With some roots, tone is not realized. Withothers, the offending segmental material is altered. We will argue that the differential behavior of variousroots, especially with respect to long vowel nuclei, motivates a coherent hierarchy of cophonologies.

4.1 CV’C Syllables and disappearing tones

As discussed in Section 2.2, when a morpheme triggering tone attaches to a word ending in a CV’C syllable,tone often fails to be realized (Can Pixabaj 2006:67; recall that [CV’C] = [CVPC] in IPA notation).

(90) a. [kar] ‘fish’b. [ın-kar] ‘my fish’ (Can Pixabaj 2006:64)

(91) a. [ka’n] ‘animal’b. [in-ka’n] ‘my animal’c. *[ın-ka’n]

(92) a. [ch’o’j] ‘fight’b. [in-ch’o’j] ‘my fight’c. *[ın-ch’o’j]

(93) a. [q’a’m] ‘staircase’b. [in-q’a’m] ‘my staircase’c. *[ın-q’a’m]

30

55

80

Inte

nsi

ty (

dB

)

80

190

300

Pit

ch (

Hz)

73ms 68ms 77ms 120ms 131ms

i n k a’ n

Time (s)0 0.5475

Figure 11: Intensity, pitch, and duration for [inka’n] ‘my animal’ (final stress, no tone)

35

It is clear that only [’C] codas block tone, and not [’C] sequences in general: tone can appear on a CV’Croot when it is followed by a vowel-initial suffix and resyllabification breaks up the final [’C] cluster.

(94) a. /wa’l-ik, H/ → [wa’.lik] ‘stopped’b. /x-at-wi’n-ik, H/ → [xat.wı’.nik] ‘you ate’

(Can Pixabaj 2006:60,203)

For a smaller set of forms ending in CV’C, tone is realized, but only along with the deletion of a final codaconsonant (Can Pixabaj 2006:57).38

(95) a. [q’u’n-iik] ‘wool coat’b. [ın-q’u’] ‘my wool coat’c. *[ın-q’u’n]

38In (95) and (96) -iik marks an inalienably possessed noun in its unpossessed form, hence the equivalent glosses.

36

(96) a. [ti’n-iik] ‘meat’b. [ın-ti’] ‘my meat’c. *[ın-ti’n]

These two subclasses are united under the generalization that final CV’C syllables cannot cooccur with tone.In the first case, this configuration is avoided by the non-realization of tone; and in the second, it is avoidedby final consonant deletion. Examples (94)–(96) also show that unlike [’C] codas, codas consisting of a singleglottal stop [’] do not block tone (see also (28)).

What about final CV’C syllables makes them incompatible with tone? Our proposal is that CV’Csyllables create a clash between stress-placement constraints and tone-placement constraints. This clash canbe resolved by failing to realize tone, or by turning the offending CV’C syllable into a CV’ syllable.

An obvious question is why CV’C syllables in particular have an adverse affect on accent placement. Thereis in fact reason to believe that CV’C syllables count as bimoraic in Uspanteko. First, like long vowels, CV’Csyllables are only found word-finally; that is, only in the position of main word stress (Can Pixabaj 2006:72–75,90). This striking parallel can be easily explained if CV’C syllables are bimoraic, and therefore subjectto the same prosodic constraints that determine the distribution of long vowels (e.g. WSP). Second, someCV’C roots alternate with CVV allomorphs (though the details need to be worked out; see Can Pixabaj2006:77).

(97) CV’C ∼ CVV alternations

a. [j-po’t=aq] ∼ [j-poo=t’aq] ‘their blouses’b. [j-to’q=aq] ∼ [j-too=t’aq] ‘their diaper’c. [r-ati’t=aq] ∼ [r-atii=t’aq] ‘their grandmother’

This pattern makes sense if (i) CV’C syllables count as heavy, and (ii) such alternations preserve the moracount of underlying forms.

If we are correct in taking CV’C syllables to be bimoraic, then they should attract stress under pressurefrom high-ranked WSP. Since CV’C syllables are restricted to word-final position, where default stress isassigned, CV’C syllables will not normally exert a visible effect on stress placement.

(98) Final CV’C inert for default stress

/ kuwa’y / All-Ft-R WSP Iamb

a. ☞ (ku.wa’y)

b. (ku.wa’y) *! *

[kuwa’y] - ‘horse’

Now consider what happens when tone is introduced. Since tone must coincide with stress in Uspanteko,we might expect stress retraction to occur. However, an iambic-trochaic reversal would leave a final CV’Cunstressed, violating high-ranked WSP. A second option would be to realize tone—and thus stress—on thefinal CV’C syllable. This solution is no better: a final CV’C syllable would have tone on the penultimatemora of the word, but on the ultimate TBU (= vocalic mora µV ), violating NonFin(T, tbu).

The third, attested option is to simply fail to realize tone, thus vacuously satisfying NonFin(T, tbu) andallowing WSP to be satisfied under default word-final stress assignment. As long as constraints enforcingtone realization are ranked relatively low (i.e. Max(T)), the desired outcome is correctly predicted.

(99) Final CV’C blocks tone realization: low-ranked Max(T)

37

/ in-ka’n, H / *Unstr-H NF(T, tbu) WSP Max(C) Max(T)

a. ☞ (in.ka’n) *

b. (ın.ka’) *!

c. (ın.ka’n) *!

d. (in.ka’n) *!

e. (ın.ka’n) *!

What about those forms that end in a CV’C syllable, but which undergo final consonant deletion when tonesurfaces? We assume that such forms belong to a distinct cophonology in which the pressure to realize inputconsonants (enforced by Max(C)) is less important than the pressure to realize tone.

(100) CV’C forms with C-deletion: Max(T) ≫ Max(C)

/ in-ti’n, H / *Unstr-H NF(T, tbu) WSP Max(T) Max(C)

a. ☞ (ın.ti’) *

b. (in.ti’n) *!

c. (ın.ti’n) *!

d. (in.tı’n) *!

e. (ın.ti’n) *!

In the basic case Max(C) outranks Max(T). Final coda consonants will thus be preserved even if it meansthe non-realization of tone. For a subset of words ending in CV’C this ranking is reversed, and final codaconsonants are elided so that tone can appear.

We thus propose an essentially phonological analysis of the unusual behavior of CV’C syllables. Onemight be tempted to propose an alternate, more phonetically-oriented analysis of these facts instead. Inmany languages, glottal stops induce non-modal phonation (usually creaky voice) on adjacent vowels (Lade-foged and Maddieson 1996:75, Gordon and Ladefoged 2001). Indeed, glottalized consonants in Uspanteko(including glottal stop) do have some effect on the phonation type of adjacent vowels, at least impressionis-tically. And as is well-known, non-modal phonation is antagonistic to the realization of high tone (Silverman1997). We might conclude, then, that CV’C syllables are incompatible with tone because the coda glottalstop gives rise to non-modal phonation on the preceding vowel, thereby interfering with its ability to expresstone.

While there is no doubt some truth to this view, there are three reasons to reject it as insufficient. First,the problem isn’t simply that CV’C syllables can’t bear tone—their presence also blocks the realization oftone on the preceding syllable (see (91)–(96)). The non-local effect that CV’C syllables exert on tone arguesagainst any analysis that depends on the local phonetic properties of those syllables themselves. Second,while codas consisting of a single glottal stop [’] also affect the phonation quality of the preceding vowel,they do not affect the realization of tone. The presence of non-modal phonation thus does not suffice toexplain the tone-blocking effect of final CV’C syllables. Finally, examples like (94) show that reference tophonological properties (i.e. syllable structure) is independently needed to explain when a [’C] sequence willinterfere with tone and when it will not. Since the phonological analysis proposed here is also sufficient toexplain the interaction of tone and CV’C syllables, it should be favored over purely phonetic alternatives.

To close, we ask why [CV’C] syllables should count as bimoraic in Uspanteko. We suspect that the glottalstop [’] in [V’C] rimes might actually be parsed as part of the syllable nucleus, creating a branching andtherefore heavy nuclear constituent. There are two reasons why this proposal is plausible. First, complexcodas are generally disallowed in Uspanteko, so assuming a nuclear parse for the glottal stop in [V’C] rimes isconsistent with broader facts about the syllable structure of the language. Second, there are many languagesin which glottal stops pattern as vowel-like or as a type of suprasegmental, especially in post-vocalic position(see e.g. various Mesoamerican examples in ?).

38

In this section we claimed that tone and final CV’C syllables exert conflicting and irreconcilable demandson metrical structure, which can only be met by eliminating tone or by breaking the [’C] coda responsiblefor the syllabic weight of CV’C. The behavior of CV’C syllables with respect to tone thus provides fur-ther evidence for a metrically-based theory of tone and accent in Uspanteko. We discuss the evidence forcophonologies in the Uspanteko noun system in more detail in the next section.

4.2 Final long vowels and more nominal cophonologies

Uspanteko nouns fall into a number of subclasses with respect to the interaction of tone and vowel length.These subclasses emerge most clearly under possession, since possession often introduces pitch accent onotherwise non-tonal nouns, thereby creating the conditions for tone-length interactions to occur.

First, we find that some nouns show vowel shortening when they appear with a possessive prefix.39

(101) a. [kaa’] ‘grinding stone’b. [ın-ka’] ‘my grinding stone’ (Can Pixabaj 2006:53)

(102) a. [sii’] ‘firewood’b. [ın-si’] ‘my firewood’ (Can Pixabaj 2006:65)

(103) a. [teem] ‘chair’b. [ın-tem] ‘my chair’ (Can Pixabaj 2006:66)

Importantly, vowel shortening is not generally required for the realization of tone: forms like *[in-kaa’] and*[in-teem] are attested elsewhere in Uspanteko (cf. [in-chaaj], [in-b’oot], etc.; see Can Pixabaj 2006:69 andSection 4.3).

Second, some nouns containing a long vowel fail to realize tone when possessed.

(104) a. [choox] ‘godmother’b. [in-choox] ‘my godmother’c. *[in-choox], *[ın-chox] (Can Pixabaj 2006:68)

(105) a. [pooq’] ‘moth’b. [in-pooq’] ‘my moth’c. *[in-pooq’], *[ın-poq’] (Can Pixabaj 2006:76)

(106) a. [keem] ‘weaving’b. [in-keem] ‘my weaving’c. *[in-keem], *[ın-kem] (Can Pixabaj 2006:68)

These two nominal subclasses can be unified under a single generalization: both types of noun avoid realizingtone on a word-final long vowel. It should be stressed that this is a parochial property of only a subset ofnouns in Uspanteko: other possessed nouns do allow tone on a final long vowel.

(107) a. [ooj] ‘avocado’b. [aw-ooj] ‘your avocado’ (Can Pixabaj 2006:93)

We suggest that Uspanteko nouns are sorted into (at least) three distinct cophonologies (Fries and Pike1949), and that the category boundaries between these cophonologies are determined by the interaction oftone and vowel length. At the coarsest level of organization, we divide Uspanteko nouns into two classes:those nouns that allow the realization of tone on a final long vowel, and those nouns that do not. We thenfurther subdivide the second class into those nouns that repair [VV#] via vowel shortening, and those thatinstead block the realization of tone. These nominal categories can be visualized as in (108).40

39There are also some nouns that lengthen under possession: for example, [k’aj] ‘wheat’ becomes [in-k’aaj] ‘my wheat’(Can Pixabaj 2006:70). Lengthening under possession is a common morphophonemic change in K’ichean-branch Mayan lan-guages (see Dayley 1985 for Tz’utujil, and Campbell 1977 on Proto-K’ichean), and is plausibly not phonological in character.

40Cophonologies of this sort generally arise as a result of language contact (e.g. Fries and Pike 1949, Ito and Mester 1995

39

(108) Structure of Uspanteko noun cophonologies

Nouns

X[VV#] *[VV#]

Vowel shortening Tone blocked

Following Anttila (2002) and Inkelas and Zoll (2007) (among others), we assume that each level of the lexiconin (108) is associated with a different ranking of constraints. At the root node of the tree, the relative rankingof at least some constraints is underspecified (e.g. A, B ≫ C). As one moves from the root node of thetree down toward its terminal nodes, the constraint set is subject to more and more specific rankings ofconstraints (e.g. A ≫ B ≫ C, B ≫ A ≫ C). (See (112) below for an instantiation of this idea.)

We propose that the dispreference for [VV#] vowels is due to the activity of another tone-related non-finality constraint, NonFin(T, σ). This constraint is simply a specific version of NonFin(T), parameterizedto the syllable rather than the TBU.41 Since long vowels are always word-final in Uspanteko, and final shortvowels never bear tone, NonFin(T, σ) will effectively penalize all and only those long vowels associatedwith pitch accent.

Nouns that allow tone on a final long vowel must be associated with a cophonology in which NonFin(T,σ) is crucially subordinated. Since both tone and underlying vowel length surface in these forms, we assumethat NonFin(T, σ) is outranked by Max(T) and Id(Length) as in tableau (109).

(109) Cophonology 1: tone allowed on long vowels; {Max(T), Id(Len)} ≫ NF(T, σ)

/ aw-ooj, H / Max(T) Id(Length) NonFin(T, σ) Iamb

a. ☞ (a.wooj) *

b. (a.woj) *! *

c. (a.wooj) *!

With NonFin(T, σ) ranked low, familiar constraints on prosodic structure come into play and locate toneon the first mora of the final long vowel as expected. For this class of nouns, then, tone on a final long vowelis tolerated in order to preserve both underlying tone and underlying vowel length.

The workings of NonFin(T, σ) can be seen more clearly in the other two noun classes, which bothdisallow tone on long vowels. For those possessed nouns that fail to realize tone when it would fall on a finallong vowel, NonFin(T, σ) and Ident(Length) must dominate Max(T).

(110) Cophonology 2: tone blocked; {NonFin(T, σ), Id(Length)} ≫ Max(T)

/ in-keem, H / NonFin(T, σ) Id(Length) Max(T) Iamb

a. ☞ (in.keem) *

b. (ın.kem) *! *

c. (in.keem) *!

With NonFin(T, σ) outranking Max(T), the non-realization of tone will be preferable to placing tone ona final long vowel. The further ranking Id(Length) ≫ Max(T) ensures that violations of NonFin(T,σ) will be repaired by the non-realization of tone rather than than by vowel shortening. In other words, forthis class of nouns it’s more important to preserve underlying vowel length than to preserve underlying H

and related work). Given the lack of large-scale historical work on Uspanteko, it is not currently possible to determine whetherthe cophonologies we propose correspond to borrowings from different source languages. However, it would be unsurprising ifthese cophonologies did have their origin in language contact: despite being geographically isolated, Uspanteko speakers havebeen in contact with speakers of K’ichee’, K’eqchi’, and the more distantly related language Ixil for a very long time.

41See Zec (2003), Flack (2009), and Gordon and Applebaum (2010) for arguments that metrical markedness constraints canbe parameterized to different levels of the prosodic hierarchy.

40

tone.This of course points the way to our next cophonology. Nouns that shorten final long vowels in order to

realize tone must have the opposite ranking of Id(Length) and Max(T).

(111) Cophonology 3: vowel shortening with tone; Max(T) ≫ Id(Length)

/ in-kaa’, H / NonFin(T, σ) Max(T) Id(Length) Iamb

a. ☞ (ın.ka’) * *

b. (in.kaa’) *!

c. (in.kaa’) *!

The ranking NonFin(T, σ) ≫ Max(T) remains unchanged for these nouns, but the new ranking Max(T)≫ Id(Length) ensures that tone on a final long vowel will be avoided via vowel shortening rather thannon-realization of tone. As opposed to the previous noun class, nouns belonging to this third cophonologypreserve underlying tone at the expense of underlying vowel length.

Though these last two noun subclasses are superfically quite distinct—one class surfaces with tone butnot long vowels, while the other class shows the converse pattern—they differ only in the relative rankingof Ident(Length) and Max(T).42 To be sure, the ranking NonFin(T, σ) ≫ Max(T) is only crucial forthose nouns that fail to realize tone (compare tableaux (110) and (111)), but we can safely assume that itholds across all nouns that disallow final long vowels bearing tone.

In (112) we provide a graphical summary of the Uspanteko noun cophonologies argued for in this section.The first cut concerns the relative ranking of Max(T) and NonFin(T, σ). Nouns that allow tone on a finallong vowel are associated with the ranking Max(T) ≫ NonFin(T, σ); nouns that do not are associatedwith the opposite ranking NonFin(T, σ) ≫ Max(T). For nouns that disallow tone on a final long vowel,either vowel shortening occurs (Max(T) ≫ Id(Length)), or tone is simply not realized (Id(Length) ≫Max(T)).

(112) Uspanteko noun cophonologies: a partial-ordering model (Anttila 2002)

Max(T) NonFin(T, σ)≫ ≫

NonFin(T, σ) Max(T)

Id(Length) Max(T) Id(Length)≫ ≫ ≫

NonFin(T, σ) Id(Length) Max(T)

X[VV#] *[VV#], shortening *[VV#], no tone

4.3 Tonal roots

The relation between tone and vowel length is somewhat more straightforward when we consider root nounsthat bear underlying accent on a final long vowel.

(113) a. [b’oot] ‘cotton’b. [kuuk’] ‘squirrel’c. [chaaj] ‘ash’ (Can Pixabaj 2006:69)

42We assume that the ranking Max(T) ≫ Iamb holds more generally in Uspanteko (see Figure 10 which shows this), sincefoot-form reversals normally occur in order to realize underlying tone on penultimate short vowels.

41

In particular, underlying tones surface faithfully when these nouns are possessed.

(114) a. [in-b’oot]b. [in-kuuk’]c. [in-chaaj] (Can Pixabaj 2006:69)

This pattern amounts to a ‘grandfather effect’ (McCarthy 2003a; Bakovic 2011): tone is permitted on a finallong vowel if it is present underlyingly, but may not be permitted if tone emerges because of morphologicalfactors like possessive marking.

Bare root nouns with tonal long vowels must belong to the cophonology in which Max(T) dominatesNonFin(T, σ) — otherwise, underlying root-specified tone would never surface in the first place.

(115) [VV#] permitted if accent is underlying: Max(T) ≫ NonFin(T, σ)

/ kuuk’, H / Max(T) NonFin(T, σ)

a. ☞ (kuuk’) *

b. (kuuk’) *!

This ranking alone is not sufficient to account for the fact that word-final accent remains in place whenthese nouns are possessed. Some additional constraint—call it Faith—must penalize candidates that avoidrealizing tone on a long vowel by way of vowel shortening.

(116) Faith protects underlying [VV#] accent

/ in-kuuk’, H / Faith Max(T) NonFin(T, σ)

a. ☞ (in.kuuk’) *

b. (in.kuuk’) *!

c. (ın.kuk’) *!

Faith could stand for any constraint that rules out vowel shortening and consequent tone shift for rootnouns with underlying [VV#] accent. For example, either Id(Length) (as in cophonology 2; see (110)) oran output-output constraint (Benua 2000) like Max-OO(T) would be sufficient for this purpose.

(117) Max-OO(T)Assign one violation for every toneless, affixed root that corresponds to a bare root form that doesbear tone.

Assuming the relevant constraint is Id(Length) would place tonal CVVC roots in the same category asnon-tonal CVVC roots that bear pitch accent under possession while also preserving vowel length (see Section4.2). This approach misses the basic generalization that CVVC roots always bear tone when possessed, whileonly some possessed CVVC roots surface with a pitch accent. So without conclusively settling the issue, weassume that Faith is just the output-output constraint Max-OO(T), thereby capturing the fact that tonalCVVC noun roots behave as a uniform class under possession.

To summarize, this section has shown that constraints on tone placement can conflict with the segmentalstructure of final syllables. Crucially, though, Uspanteko does not make use of a uniform strategy to resolvethese conflicts. Some roots block tone realization, while other roots are unfaithfully realized in order toaccomodate tone. Our main proposal is that Uspanteko nouns are arranged into cophonologies that have ahierarchical structure generated by reranking only pairs of constraints.

This concludes our analysis of accent and prosodic structure in Uspanteko. We have shown that a metrical,foot-based analysis of word-level prosody in Uspanteko not only accounts for the basic distribution of tone,but also makes sense of several interactions between accent and segmental structure. The non-local characterof some of these interactions provides further support for a foot-based treatment of Uspanteko phonology,a point made explicitly in Section 3.3. Finally, we discussed lexical variation in the interaction of tone andvowel length in Uspanteko, and provided an analysis of possessive inflection in terms of a partial-orderingmodel of phonological organization (e.g. Anttila 2002).

42

5 Conclusion

This paper presents the first formal analysis of Uspanteko accent. The accentual system of Uspanteko isan important area of study, not just because the language is threatened and understudied, but because itcontributes valuable data to the typology of accent systems. Uspanteko makes use of two distinct non-iterative accent systems. Unlike other accent systems that combine stress and tone (e.g. Stockholm Swedish,Bruce 1977, Gussenhoven 2004:210; Ayutla Mixtec, de Lacy 2002; Somali nouns, Hyman 1981, Gussenhoven2004:39), non-tonal words are permitted in Uspanteko. Unstressed words, on the other hand, are notpermitted. Uspanteko thus simultaneously instantiates an obligatory accent system (stress) and a non-obligatory accent system (pitch accent) (Hyman 2006).

Much work on the typology of such ‘hybrid’ accent systems makes a three-way distinction between lan-guages in which stress placement determines tone placement, languages in which tone placement determinesstress placement, and languages in which the two types of accent do not interact (e.g. van der Hulst andSmith 1988, ?:250-1, ?). Uspanteko instantiates a fourth, alternative category: stress placement and toneplacement are co-determined. While tone is attracted to the position of stress, constraints on the distributionof tone (e.g. the NonFinality(T) family) can cause tone to displace rightward, bringing stress along withit. There is no sense in which stress is ‘prior’ to pitch accent, or vice versa; they each exert independent,but interrelated demands which must be simultaneously met. (See Michael 2010 for related discussion oftone-stress interactions in Iquito.)

We have also shown that both tone and stress in Uspanteko are subject to strict metrical constraints,including constraints on foot structure. In some ways, this is a very surprising result: neither stress orpitch accent are ‘rhythmic’ in Uspanteko (i.e. there are no secondary or alternating accents); and word-level accent, which is highly regular, could be easily described without any reference to the foot at all.Nevertheless, there is strong evidence that Uspanteko words contain a single, right-aligned foot that governsboth accent placement as well as interactions between prosodic and segmental structure. This paper thusadds to a growing body of literature suggesting that a small set of general prosodic categories are universallyinstantiated in every language (see e.g. Hayes 1995:119, Jun 2005, Goad and Buckley 2006, Kawahara andShinya 2008, Ito and Mester 2009, Vogel 2009, Selkirk 2011, and references therein for discussion).

Even though accent in Uspanteko is determined by sharp restrictions on prosodic structure, differentlexical items show different behavior in coping with these constraints. We saw that tone and vowel lengthinteract in a non-uniform way in Uspanteko words. This otherwise intractable data can be elegantly capturedusing cophonologies, defined by minimal, pairwise re-ranking of constraints for different subsets of the lexicon.

While this paper provided an analysis of the core facts regarding word-level prosody in Uspanteko, manyinteresting questions remain. One important issue regards the phonetic realization of stress in Uspanteko,given that stress placement is highly predictable, and common phonetic cues to stress (i.e. vowel length andpitch) are used contrastively in the language (see Berinstein 1979). A related question is whether cues tostress in Uspanteko are identical in tonal and non-tonal syllables. We would also like to better understandthe conditions on syncope sketched in Section 3.3.2, as our suspicion is that syncope may be blocked inparticular phrasal contexts. Consequently, we would also like to better understand how word-level prosodyin Uspanteko is integrated into higher prosodic structure.

Acknowledgments: Above all we thank our primary Tz’unun Kaab’ (Uspanteko) consultants Juana AjpoopTikiram and Miguel Pinula Pacheco, as well as Juan Us and the rest of the Comunidad Linguıstica Uspanteka.This paper was substantially improved by feedback from Judith Aissen, Melissa Frazier, Maria Gouskova,Larry Hyman, Junko Ito, Armin Mester, and the participants in the Winter 2010 Pitch Accent seminar atUC Santa Cruz. We are also grateful to audiences at the 2010 International Symposium on Accent andTone in Tokyo, the CUNY Conference on the Phonology of Endangered Languages, and the UC BerkeleyFieldwork Forum for comments on this work. Finally, we thank three anonymous reviewers for helping usto strengthen our argumentation. The authors’ names appear in alphabetical order.

References

Aissen, Judith. 1999. External possessor and logical subject in Tz’utujil. In External possession, eds. DorisPayne and Immanuel Barshi. Amsterdam: John Benjamins.

43

Anttila, Arto. 2002. Morphologically conditioned phonological alternations. Natural Language and LinguisticTheory 20 (1): 1–42.

Bakovic, Eric. 2011. Opacity and ordering, 2nd edn. In The handbook of phonological theory, eds. JohnGoldsmith, Alan C. L. Yu, and Jason Riggle, 40–67. Malden MA: Wiley-Blackwell.

Barnes, Jonathan. 2006. Strength and weakness at the interface: positional neutralization in phonetics andphonology. Berlin: Mouton de Gruyter.

Beckman, Jill. 1998. Positional faithfulness. PhD diss, University of Massachusetts, Amherst.

Benua, Laura. 2000. Phonological relations between words. New York: Garland.

Berinstein, Ava E. 1979. A cross-linguistic study on the perception and production of stress. Master’s thesis,University of California Los Angeles. Available online at http://escholarship.org/uc/item/0t0699hc.

Bishop, Judith. 2002. Stress accent without phonetic stress: accent type and distribution in Bininj Gun-Wok.In Proceedings of the Speech Prosody 2002 conference, eds. Bernard Bel and Isabelle Marlien, 179–182.

Blumenfeld, Lev. 2006. Constraints on phonological interactions. PhD diss, Stanford University.

Bollinger, Dwight. 1958. A theory of pitch acent in English. Word 14: 109–149.

Bruce, Gosta. 1977. Swedish word accents in sentence perspective. LiberLaromedel/Gleerup.

Buckley, Eugene. 1998. Iambic lengthening and final vowels. International Journal of American Linguistics64 (3): 179–223.

Cairns, Charles. 2002. Foot and syllable in Southern Paiute City University of New York.

Campbell, Lyle. 1977. Quichean linguistic prehistory. Vol. 81 of University of california publications inlinguistics. Berkeley, CA: University of California Press.

Campos-Astorkiza, Rebeka. 2007. Minimal contrast and the phonology-phonetics interaction. PhD diss,University of Southern California.

Can Pixabaj, Telma. 2006. Gramatica descriptiva Uspanteka. Oxlajuuj Keej Maya’ Ajtz’iib’ (OKMA).

Cassimjee, Farida, and Charles Kisseberth. 2007. Optimal domains theory and Bantu tonology: a casestudy from Isixhosa and Shingasidja. In Theoretical aspects of Bantu tone, eds. Larry Hyman and CharlesKisseberth, 33–132. Stanford, CA: CSLI Publications.

Chomsky, Noam. 2001. Derivation by phase. In Ken Hale: A life in language, ed. Michael Kenstowicz, 1–52.Cambridge, MA: MIT Press.

Clements, G. N. 1990. The role of the sonority cycle in core syllabification. In Papers in laboratory phonologyI: Between the grammar and the physics of speech, eds. John Kingston and Mary Beckman, 283–333.Cambridge, UK: Cambridge University Press.

Crowhurst, Megan, and Lev Michael. 2005. Iterative footing and prominence-driven stress in Nanti (Kampa).Language 81 (1): 47–95.

Cutler, Anne. 2005. Lexical stress. In The handbook of speech perception, eds. David Pisoni and RobertRemez, 264–289. Malden, MA: Blackwell.

Dayley, Jon. 1985. Tzutujil grammar. Vol. 107 of University of california publications in linguistics. Berkeley,CA: University of California Press.

de Lacy, Paul. 2002. The interaction of tone and stress in Optimality Theory. Phonology 19 (1): 1–32.

de Lacy, Paul. 2004. Markedness conflation in Optimality Theory. Phonology 21 (2): 145–199.

44

de Lacy, Paul. 2007. The interaction of tone, sonority, and prosodic structure. In The Cambridge handbookof phonology, ed. Paul de Lacy, 281–307. Cambridge, UK: Cambridge University Press.

Dell, Francois. 1995. Consonant clusters and phonological syllables in French. Lingua 95 (1-3): 5–26.

Dell, Francois, and Mohamed Elmedlaoui. 1985. Syllabic consonants and syllabification in Imdlawn TashlhiytBerber. Journal of African Languages and Linguistics 7 (2): 105–130.

Demers, Richard, Fernando Escalante, and Eloise Jelinek. 1999. Prominence in Yaqui words. InternationalJournal of American Linguistics 65 (1): 40–55.

Elias-Ulloa, Jose. 2006. Theoretical aspects of Panoan metrical phonology: Disyllabic footing and contextualsyllable weight. PhD diss, Rutgers University.

Embick, David, and Morris Halle. 2005. On the status of stems in morphological theory. Romance languagesand linguistic theory 2003: selected papers from “Going Romance” 2003, Nijmegen, 20-22 November.

Embick, David, and Rolf Noyer. 2007. Distributed Morphology and the syntax/morphology interface. In TheOxford handbook of linguistic interfaces, eds. Gillian Ramchand and Charles Reiss, 289–324. Oxford, UK:Oxford University Press.

Flack, Kathryn. 2009. Constraints on onsets and codas of words and phrases. Phonology 26 (2): 269–302.

Fries, Charles, and Kenneth Pike. 1949. Coexistent phonemic systems. Language 25 (1): 29–50.

Fry, Dennis. 1955. Duration and intensity as physical correlates of linguistic stress. The Journal of theAcoustical Society of America 27 (4): 765–768.

Goad, Heather, and Meaghen Buckley. 2006. Prosodic structure in child French: Evidence for the foot.Catalan Journal of Linguistics 5: 109–142.

Gordon, Matthew. 2000. The tonal basis of final weight criteria. In Chicago Linguistics Society (CLS) 36,eds. Akira Okrent and John P. Boyle, 141–156. Chicago Linguistic Society.

Gordon, Matthew, and Ayla Applebaum. 2010. Prosodic fusion and minimality in Kabardian. Phonology 27(1): 45–76.

Gordon, Matthew, and Peter Ladefoged. 2001. Phonation types: a cross-linguistic overview. Journal ofPhonetics 29 (4): 383–406.

Gouskova, Maria. 2003. Deriving economy: syncope in Optimality Theory. PhD diss, University of Mas-sachusetts Amherst.

Green, Thomas, and Michael Kenstowicz. 1995. The Lapse constraint. Ms., MIT. Available online as ROA-101, Rutgers Optimality Archive, http://roa.rutgers.edu/.

Grimes, James. 1971. A reclassification of the Quichean and Kekchian (Mayan) languages. InternationalJournal of American Linguistics 37 (1): 15–19.

Gussenhoven, Carlos. 2004. The phonology of tone and intonation. Cambridge, UK: Cambridge UniversityPress.

Halle, Morris, and Alec Marantz. 1993. Distributed morphology and the pieces of inflection. In The viewfrom Building 20, 111–176. MIT Press.

Halle, Morris, and Alec Marantz. 1994. Some key features of Distributed Morphology. In MIT working papersin linguistics, Vol. 21, 275–288. MITWPL.

Harley, Heidi, and Rolf Noyer. 1999. State-of-the-article: Distributed morphology. Glot International 4 (4):3–9.

45

Hayes, Bruce. 1981. A metrical theory of stress rules. Bloomington, Indiana: Distributed by Indiana Univer-sity Linguistics Club. Revised version of 1980 MIT Ph.D. thesis.

Hayes, Bruce. 1995. Metrical stress theory. The University of Chicago Press.

Hayes, Bruce, and Aditi Lahiri. 1991. Bengali intonational phonology. Natural Language and LinguisticTheory 9 (1): 47–96.

Hockett, Charles. 1966. The problem of universals in language, 2nd edn. In Universals of language, ed. JosephGreenberg, 1–29. Cambridge, MA: MIT Press.

Hyman, Larry. 1977. On the nature of linguistic stress. In Studies in stress and accent, ed. Larry Hyman. Vol.4 of Southern california occasional papers in linguistics, 37–82. Los Angeles: Department of Linguistics,University of Southern California.

Hyman, Larry. 1981. Tonal accent in Somali. Studies in African linguistics 12 (2): 169–203.

Hyman, Larry. 2006. Word-prosodic typology. Phonology 23 (2): 225–257.

Hyman, Larry. 2007. Universals of tone rules: 30 years later. In Tones and tunes: typological studies inword and sentence prosody, eds. Tomas Riad and Carlos Gussenhoven, Vol. 1, 1–34. Berlin: Mouton DeGruyter.

Icke, Ignacio. 2007. Map of Guatemalan languages. Available online athttp://en.wikipedia.org/wiki/File:Idiomasmap.svg#metadata. Licensed under the GNU Free Documen-tation License, Version 1.2 and the Creative Commons Attribution-Share Alike License, Version 3.0 Un-ported.

Inkelas, Sharon. 1990. Prosodic constituency in the lexicon. New York: Garland.

Inkelas, Sharon, and Cheryl Zoll. 2007. Is grammar dependence real? A comparison between cophonologicaland indexed constraint approaches to morphologically conditioned phonology. Linguistics 45 (1): 133–171.

Ito, Junko, and Armin Mester. 1995. The core-periphery structure of the lexicon and constraints on reranking.In Papers in Optimality Theory, eds. Jill Beckman, LauraWalsh Dickey, and Suzanne Urbanczyk, 181–209.Amherst, MA: GLSA Publications.

Ito, Junko, and Armin Mester. 2007. Prosodic adjunction in Japanese compounds. In Formal approaches toJapanese linguistics (FAJL) 4, eds. Yoichi Miyamoto and Masao Ochi, 97–111. Cambridge, MA: MITWPL.

Ito, Junko, and Armin Mester. 2009. The extended prosodic word. In Phonological domains: Universals anddeviations, eds. Janet Grijzenhout and Baris Kabak, 135–194. Berlin: Mouton de Gruyter.

Ito, Junko, and Armin Mester. 2011. The perfect prosodic word and the sources of unaccentedness. Ms.,University of California, Santa Cruz.

Jespersen, Otto. 1904. Lehrbuch der phonetik. Leipzig: B.G. Teubner.

Jun, Sun-Ah. 2005. Prosodic typology. In Prosodic typology: The phonology of intonation and phrasing, ed.Sun-Ah Jun, 430–458. Oxford, UK: Oxford University Press.

Kager, Rene. 1993a. Alternatives to the iambic-trochaic law. Natural Language and Linguistic Theory 11(3): 381–432.

Kager, Rene. 1993b. Shapes of the generalized trochee. In West Coast Conference on Formal Linguistics(WCCFL) 11, ed. Jonathan Mead, 298–312. Stanford, CA: CSLI Publications.

Kager, Rene. 1997. Rhythmic vowel deletion in Optimality Theory. In Derivations and constraints in phonol-ogy, ed. Iggy Roca, 463–499. Oxford: Oxford University Press.

Kager, Rene. 1999. Optimality theory. Cambridge University Press.

46

Kager, Rene. 2001. Rhythmic directionality by positional licensing. Handout at Fifth HIL Phonology Con-ference, University of Potsdam.

Kager, Rene. 2005. Rhythmic licensing theory: an extended typology. In Proceedings of the third interna-tional conference on phonology, 5–31. Seoul. The Phonology-Morphology Circle of Korea. The Phonology-Morphology Circle of Korea.

Kaufman, Terrence. 1974. Idiomas de Mesoamerica. Seminario de Integracion Social Guatemalteca.

Kaufman, Terrence. 1990. Algunos rasgos estructurales de los idiomas mayances con referencia especial alK’iche’. In Lecturas sobre la linguıstica maya, eds. Nora England and Stephen Elliott, 59–114. La Antigua,Guatemala: Centro de Investigaciones Regionales de Mesoamerica.

Kawahara, Shigeto, and Takahito Shinya. 2008. The intonation of gapping and coordination in Japanese:Evidence for Intonational Phrase and Utterance. Phonetica 65 (1-2): 62–105.

Kenstowicz, Michael. 1994. Sonority-driven stress MIT. Available online as ROA-33, Rutgers OptimalityArchive, http://roa.rutgers.edu/.

Kitagawa, Yoshihisa. 1986. Subjects in Japanese and English. PhD diss, University of Massachusetts,Amherst.

Kubozono, Haruo. 2008. Japanese accent. In The handbook of Japanese linguistics, eds. Shigeru Miyagawaand Mamoru Saito, 165–191. New York: Oxford University Press.

Ladefoged, Peter, and Ian Maddieson. 1996. The sounds of the world’s languages. Malden, MA: Blackwell.

Liberman, Mark. 1975. The intonational system of English. PhD diss, Massachusetts Institute of Technology.

Liberman, Mark, and Alan Prince. 1977. On stress and linguistic rhythm. Linguistic Inquiry 8 (2): 249–336.

Manzini, Rita. 1983. Restructuring and reanalysis. PhD diss, Massachusetts Institute of Technology.

Marantz, Alec. 1991/2000. Case and licensing. In Arguments and case: Explaining Burzio’s generalization,ed. Eric Reuland, 11–30. Amsterdam: John Benjamins. Originally appeared 1991 in Proceedings of theEastern States Conference on Linguistics (ESCOL) 8, Ohio State University Department of Linguistics.

McCarthy, John. 1986. OCP effects: gemination and antigemination. Linguistic Inquiry 17 (2): 207–263.

McCarthy, John J. 2003a. Comparative markedness. Theoretical Linguistics 29 (1-2): 1–51.

McCarthy, John J. 2003b. OT constraints are categorical. Phonology 20: 75–138.

McCarthy, John J. 2008. The serial interaction of stress and syncope. Natural Language and LinguisticTheory 26 (3): 499–546.

McCarthy, John J., and Alan Prince. 1993. Generalized alignment. Yearbook of Morphology. Available onlineas ROA-7, Rutgers Optimality Archive, http://roa.rutgers.edu/.

McCarthy, John J., and Alan Prince. 1994. The Emergence of the Unmarked: Optimality in prosodic mor-phology. In Proceedings of NELS, ed. Merce Gonzalez, Vol. 24, 333–379.

McCarthy, John J., and Alan Prince. 1995. Faithfulness and reduplicative identity. In University of Mas-sachusetts occasional papers in linguistics, eds. Jill Beckman, LauraWalsh Dickey, and Suzanne Urbanczyk,249–384.

Mendez, Miguel Angel Vicente. 2007. Diccionario bilingue uspanteko-espanol. Oxlajuuj Keej Maya’ Ajtz’iib’.

Mester, Armin. 1994. The quantitative trochee in Latin. Natural Language and Linguistic Theory 12 (1):1–61.

47

Michael, Lev. 2010. The interaction of tone and stress in the prosodic system of Iquito (Zaparoan). In UCBerkeley Phonology Lab annual report, 57–79. Department of Linguistics, UC Berkeley. Available online athttp://linguistics.berkeley.edu/phonlab/annual report/documents/2010/Iquito prosody BRILL v8.pdf.

Myers, Scott. 1998. Surface underspecification of tone in Chichewa. Phonology 15 (3): 367–391.

Myers, Scott, and Benjamin Hansen. 2007. The origin of vowel length neutralization in final position: Evi-dence from Finnish speakers. Natural language and linguistic theory 25 (1): 157–193.

Nelson, Nicole. 2003. Asymmetric anchoring. PhD diss, Rutgers University.

Nevins, Andrew. 2007. The representation of third person and its consequences for person-case effects. NaturalLanguage and Linguistic Theory 25 (2): 273–313.

Pater, Joe. 2000. Non-uniformity in English secondary stress: the role of ranked and lexically specific con-straints. Phonology 17 (2): 237–274.

Prince, Alan. 1991. Quantitative consequences of rhythmic organization. In Cls 26(2): Papers from theparasession on the syllable in phonetics and phonology, eds. Karen Deaton, Manuela Noske, and MichaelZiolkowski, 355–398. Chicago: Chicago Linguistics Society.

Prince, Alan, and Paul Smolensky. 1993/2004.Optimality Theory: constraint interaction in generative gram-mar. Malden, MA: Blackwell. Revision of 1993 technical report, Rutgers University Center for CognitiveScience. Available online as ROA-537, Rutgers Optimality Archive, http://roa.rutgers.edu/.

Pruitt, Kathryn. 2010. Serialism and locality in constraint-based metrical parsing. Phonology 27 (3): 481–526.

Pulleyblank, Douglas. 1986. Tone in lexical phonology. Dordrecht: D. Reidel.

Pye, Clifton. 1983. Mayan telegraphese: Intonational determinants of inflectional development in QuicheMayan. Language 59 (3): 583–604.

Richards, Michael. 2003. Atlas linguıstico de Guatemala. Instituto de Linguıstico y Educacion de la Univer-sidad Rafael Landıvar.

Selkirk, Elisabeth. 2011. The syntax-phonology interface, 2nd edn. In Handbook of phonological theory, eds.John Goldsmith, Alan C. L. Yu, and Jason Riggle. Blackwell handbooks in linguistics series, 435–484.Malden MA: Wiley-Blackwell.

Silverman, Daniel. 1997. Laryngeal complexity in Otomanguean vowels. Phonology 14 (2): 235–261.

Silverman, Kim, and Janet Pierrehumbert. 1990. The timing of pre-nuclear high accents in English. In Papersin laboratory phonology I: Between the grammar and the physics of speech, eds. John Kingston and MaryBeckman, 72–106. Cambridge, UK: Cambridge University Press.

Smith, Jennifer. 2005. Phonological augmentation in prominent positions. Routledge.

Steriade, Donca. 2000. Paradigm uniformity and the phonetics-phonology boundary. In Papers in laboratoryphonology v: Acquisition and the lexicon, eds. Michael Broe and Janet Pierrehumbert, 313–334. Cambridge,UK: Cambridge University Press.

Teeple, David. 2009. Biconditional prominence correlation. PhD diss, University of California, Santa Cruz.

Trubetzkoy, Nikolai. 1939. Grundzuge der phonologie. Travaux du cercle linguistique de Prague. Englishtranslation published 1969 as Principles of phonology, trans. C.A.M. Baltaxe. Berkeley: University ofCalifornia Press.

van de Vijver, Ruben. 1998. The iambic issue: iambs as a result of constraint interaction. The Hague:Holland Academic Graphics.

48

van der Hulst, Harry, and Norval Smith. 1988. The variety of pitch accent systems: Introduction. Dordrecht:Foris Publications.

Vogel, Irene. 2009. Universals of prosodic structure. In Universals of language today, eds. Sergio Scalise,Elisabetta Magni, and Antonietta Bisetto, 59–82. Springer.

Woolford, Ellen. 1997. Four-way case systems: Ergative, nominative, objective and accusative. NaturalLanguage and Linguistic Theory 15 (1): 181–227.

Yip, Moira. 2002. Tone. Cambridge, UK: Cambridge University Press.

Zec, Draga. 1999. Footed tones and tonal feet: rhythmic constituency in a pitch-accent language. Phonology16 (2): 225–264.

Zec, Draga. 2003. Prosodic weight. In The syllable in optimality theory, eds. Caroline Fery and Ruben vande Vijver, 123–143. Cambridge, UK: Cambridge University Press.

Appendix: domain effects

Up till this point we have assumed that, in Uspanteko, metrical structure is built with respect to the rightedge of the prosodic word (ω). This view is complicated by certain prosodic effects found with cliticization.Uspanteko has a number of enclitics (at least emphatic i’(n) and plural aq) that disrupt otherwise excep-tionless phonological generalizations (Can Pixabaj 2006:52–3). As discussed in Section 3, long vowels arerestricted to final position within the word.

(118) a. chuun ‘lime (mineral)’b. x-chun-aj ‘he covered it with lime’ (Can Pixabaj 2006:53)

(119) a. in-jii’ ‘my son-in-law’b. ji’-xeel ‘son-in-law’c. ji’-xel-ib’ ‘sons-in-law’ (Can Pixabaj 2006:62,69)

However, unlike true suffixes, enclitics fail to trigger shortening of final long vowels.

(120) a. tz’eet ‘true’b. tz’eet=i’ ‘it is true’ (Can Pixabaj 2006:52,667)

(121) a. poot’ ‘blouse’b. j-poot’=aq ‘their blouses’ (Can Pixabaj 2006:53,76)

Similarly, tone is normally restricted to the penultimate mora (Can Pixabaj 2006:62-69,etc.).

(122) a. ın-pix ‘my tomato’b. *in-pıx (Can Pixabaj 2006:64)

(123) a. kuuk’ ‘squirrel’b. *kuuk’ (Can Pixabaj 2006:22)

(124) a. ji’-xel-ib’ ‘sons-in-law’b. *ji’-xel-ıb’c. *jı’-xel-ib’ (Can Pixabaj 2006:62)

But when enclitics appear, they fail to trigger rightward tone shift.

(125) a. ıxim ‘corn’b. w-ıxim=aq ‘my corn (pl.)’c. *w-ixım=aq

49

30

55

80

Inte

nsi

ty (

dB

)

80

190

300

Pit

ch (

Hz)

118ms 94ms 166ms 343ms

j p i x

Time (s)0 0.9765

Figure 12: Intensity, pitch, and duration for [j-pix] ‘his/her tomato’

Finally, even in toneless forms enclitics do not affect the position of accent on their preceding hosts. AsFigures 12 and 13 illustrate, final stress (as cued by phonetic vowel lengthening) stays in place when followedby the plural enclitic aq.We conclude from these facts that the building of metrical structure, and thus the assignment of tone andstress accent, occurs within the minimal prosodic word (ωmin) in Uspanteko (e.g. Inkelas 1990; Ito andMester 2007, 2009). On the assumption that enclitics like i’(n) and aq adjoin to the minimal prosodic word,the prosodic structure of their hosts is correctly predicted to remain unaltered by encliticization. Given thevolume and frequency of clitics in Uspanteko (and in Mayan languages more generally), there is no doubtmore to say about the prosodic phonology of cliticization; for now, we leave these questions open for futureresearch.

50

30

55

80

Inte

nsi

ty (

dB

)

80

190

300

Pit

ch (

Hz)

113ms 58ms 125ms 153ms 80ms 304ms

j p i x a q

Time (s)0 0.9402

Figure 13: Intensity, pitch, and duration for [j-pix=aq] ‘their tomato’

51